Chart data

2391 charts · 73,917 points · 12 labs

Every chart, as data.

The numbers behind the charts the labs publish, with where each one came from. Study them, redraw them, or make your own. Charts marked Replica have a look-alike redraw that scored 7 or more out of 10.

Use it as files Pricing

Charts2391

Points73,917

Publisher's numbers31fully or partly

Estimated from images2360

Labs12

Accepted replicas95look-alike score 7 or more of 10

OpenAI 211

Estimated from the imageReplica 7.4TerminalBench 2.1OpenAI · Bar chart · 9 points Estimated from the imageReplica 8.1Illustrative global managed assetsOpenAI · Grouped bar chart · 22 points Estimated from the imageReplica 8.1Illustrative global managed assetsOpenAI · Grouped bar chart · 22 points Partly publisher's numbersReplica 8.3Accuracy / Percentile (%)deepseek · Benchmark bar chart · 30 points Estimated from the imageModel Benchmark ComparisonOpenAI · Table chart · 42 points Estimated from the imageGPT4 Turbo Benchmark ComparisonOpenAI · Line chart · 12 points Estimated from the imageGrok-4.20-Beta1 ranks #1 in Search Arenaarena · Table chart · 40 points Estimated from the imageBenchmark ComparisonOpenAI · Table chart · 27 points Estimated from the imageArtificial Analysis Intelligence Index v4.1.1artificial-analysis · Bar chart · 29 points Estimated from the imageArtificial Analysis Intelligence Index v4.1OpenAI · Scatter plot · 14 points Estimated from the imageReplica 8.1Artificial Analysis Intelligence Index v4.1artificial-analysis · Grouped bar chart · 30 points Estimated from the imageReplica 8.9Artificial Analysis Intelligence Index v4.1artificial-analysis · Bar chart · 44 points Estimated from the imageAgent Performance vs Humans with Time Limits (CI 95%)OpenAI · Bar chart · 6 points Estimated from the imageNeurons spike rate, Paw movement, Licking, Mouth movement, Run speed, Pupil diameter, global astrocytic signalOpenAI · Line chart · 40 points Estimated from the imageReplica 9.2Illustrative global managed assets have more than doubled since 2015OpenAI · Grouped bar chart · 22 points Estimated from the imageExploitBenchOpenAI · Line chart · 22 points Estimated from the imageFrontierScience-Research accuracy across modelsOpenAI · Bar chart · 8 points Estimated from the imageReplica 5.7Change in combined output tokens by department since November 2025OpenAI · Line chart · 20 points Estimated from the imageOpenAI o3-pro: comparative evaluations with human testersOpenAI · Bar chart · 5 points Estimated from the imageModel Performance ComparisonOpenAI · Bar chart · 4 points Estimated from the imageAI Model Benchmark ComparisonOpenAI · Table chart · 30 points Estimated from the imageAI Model Benchmark ComparisonOpenAI · Table chart · 31 points Estimated from the imageVisual-6 Luna (Max) · RefCOCOgOpenAI · Bar chart · 2 points Estimated from the imageArtificial Analysis Coding IndexOpenAI · Line chart · 34 points Estimated from the imageFrontierScience-Olympiad accuracy across modelsOpenAI · Bar chart · 8 points Estimated from the imageAIME 2025 Competition mathOpenAI · Benchmark bar chart · 9 points Estimated from the imageHealthBench ProfessionalOpenAI · Line chart · 54 points Estimated from the imageBrowseCompOpenAI · Bar chart · 6 points Estimated from the imageInternal instruction following evalOpenAI · Bar chart · 2 points Estimated from the imageReplica 8.1CyberGymOpenAI · Bar chart · 6 points Estimated from the imageSWE-bench Verified Software engineeringOpenAI · Line chart · 7 points Estimated from the imageGPT-4 Pricing Over TimeOpenAI · Line chart · 4 points Estimated from the imageGDPval Knowledge work tasksOpenAI · Stacked bar chart · 6 points Estimated from the imageAccuracyOpenAI · Benchmark bar chart · 20 points Estimated from the imageGPT-4o mini is more than 60% cheaper than GPT-3.5 TurboOpenAI · Line chart · 10 points Estimated from the imageGPT-Rosalind improves core biology and chemistry benchmarksOpenAI · Benchmark bar chart · 20 points Estimated from the imageRecent released open weight models compare to frontier modelsOpenAI · Line chart · 60 points Estimated from the imageSWE-Bench Verified Software EngineeringOpenAI · Bar chart · 4 points Estimated from the imageNon-developerOpenAI · Line chart · 33 points Estimated from the imageUp to 1.5MOpenAI · Bar chart · 4 points Estimated from the imageOccupation vs. work tokens done with CodexOpenAI · Table chart · 25 points Estimated from the imageInternal hallucination with noise evalOpenAI · Bar chart · 4 points Estimated from the imageCommon VoiceOpenAI · Bar chart · 2 points Estimated from the imageDeveloperOpenAI · Line chart · 33 points Estimated from the imageOpenAI o3: 80% price reduction over previous o3OpenAI · Grouped bar chart · 6 points Estimated from the imageArtificial Analysis Coding Agent Index v1.1OpenAI · Scatter plot · 31 points Estimated from the imageGPT-5.6 Sol on the ARC-AGI-3 Public SetOpenAI · Line chart · 10 points Estimated from the imageCoding deception (lower is better)OpenAI · Bar chart · 5 points Estimated from the imageGPQA DiamondOpenAI · Grouped bar chart · 7 points Estimated from the imageNet Improvement vs. median cost per session (7-day window)arena · Scatter plot · 15 points Estimated from the imageImage Edit Arena-2.5: Ranked #1&#2arena · Bar chart · 17 points Estimated from the imageMATH Level 5 accuracy for o3 and o4-mini compared to leading modelsepoch-ai · Bar chart · 8 points Estimated from the imageGPT-Live-1 Performance BenchOpenAI · Benchmark bar chart · 6 points Estimated from the imageCode GPT-6 Arena: WebDev (Max): Rank #24arena · Bar chart · 24 points Estimated from the imageGPT-6Sol(Max) Pareto Frontier for Code Arenaarena · Line chart · 9 points Estimated from the imageSWE-Lancer IC SWEOpenAI · Bar chart · 2 points Estimated from the imageOpenAI o3 ranks #5 in WebDev Arena!arena · Table chart · 33 points Estimated from the imageArtificial Analysis Hallucination Rateartificial-analysis · Bar chart · 40 points Estimated from the imageReviews with high scores by evaluation criteriaOpenAI · Benchmark bar chart · 20 points Estimated from the imageLifeSciBench Scores by Scientific WorkflowOpenAI · Benchmark bar chart · 18 points Estimated from the imageShare of tasks Agent solved on a given attempt (pass@1) vs pass@5 (at least once) vs pass^5, on all 5 attempts (pass^5) - Top 10 models by pass^5 - Higher is betterartificial-analysis · Benchmark bar chart · 36 points Estimated from the imageSWE-bench VerifiedOpenAI · Bar chart · 2 points Estimated from the imageSWE-Bench ProOpenAI · Bar chart · 3 points Estimated from the imageCommon VoiceOpenAI · Bar chart · 2 points Estimated from the imageArtificial Analysis Speech to Speech Index by Componentartificial-analysis · Benchmark bar chart · 100 points Estimated from the imageOpenAI o3-pro: 87% price reduction compared to o1-proOpenAI · Grouped bar chart · 8 points Estimated from the imageArtificial Analysis Coding Agent Indexartificial-analysis · Bar chart · 15 points Estimated from the imageGPT-5.5 Instant: better at visual reasoningOpenAI · Benchmark bar chart · 4 points Estimated from the imageArtificial Conversational DynamicsOpenAI · Bar chart · 3 points Estimated from the imageCompetition Math (AIME 2024)arena · Bar chart · 6 points Estimated from the imageIntelligence vs. Price: Non-Reasoning Modelsartificial-analysis · Scatter plot · 14 points Estimated from the imageo3 #1 for Style Control, Coding, Hard o3 & o4 mini #1 for Matharena · Table chart · 130 points Estimated from the imageGPT-5.6 Sol and Luna (xHigh)arena · Scatter plot · 6 points Estimated from the imageGPT-5.5 Instant: a step up in STEMOpenAI · Benchmark bar chart · 4 points Estimated from the imageNano Banana 2.1: Gap to Use Case Frontier (AA-Image-T2I v2.0)artificial-analysis · Table chart · 30 points Estimated from the imageTerminal-Bench 4.0OpenAI · Line chart · 15 points Estimated from the imageCommon VoiceOpenAI · Bar chart · 4 points Estimated from the imageGPT-6.1-Sol (Max) Pareto Frontier for Code Arenaarena · Line chart · 6 points Estimated from the imageAutomationBenchOpenAI · Line chart · 32 points Estimated from the imageDeepSWEOpenAI · Line chart · 84 points Estimated from the imageFraction of Model A Wins for All Non-tied A vs. B Battlesarena · Table chart · 81 points Estimated from the imageSWE-Bench Pro (Public)OpenAI · Line chart · 15 points Estimated from the imageContext Aware ASR BenchmarkOpenAI · Benchmark bar chart · 4 points Estimated from the imageAverage time to First audio (seconds) on Big Bench Audio dataset - Lower is betterartificial-analysis · Bar chart · 26 points Estimated from the imageFrontierMath accuracyepoch-ai · Scatter plot · 6 points Estimated from the imageOutput Tokens used Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 42 points Estimated from the imageArtificial Analysis Quality Indexartificial-analysis · Bar chart · 18 points Estimated from the imageFrontierMath private accuracy for OpenAI's o3 and o4-mini by reasoning settingepoch-ai · Bar chart · 6 points Estimated from the imageArtificial Analysisartificial-analysis · Bar chart · 36 points Estimated from the imageDeepSWEOpenAI · Line chart · 16 points Estimated from the imageGPT Image 2.5 Sunburst (max): Gap to Use Case Frontier (Image Editing)artificial-analysis · Table chart · 20 points Estimated from the imageImage-to-WebDev: Ranked #1arena · Benchmark bar chart · 15 points Estimated from the imageCode Arenaarena · Bar chart · 13 points Estimated from the imageSWE-bench Verified accuracyOpenAI · Bar chart · 7 points Estimated from the imageArtificial Analysis Image Arena Leaderboard (Text to Image)artificial-analysis · Table chart · 10 points Estimated from the imageCost (USD) to Run all Evaluations in the Artificial Intelligence Indexartificial-analysis · Benchmark bar chart · 31 points Estimated from the imageOutput Speedartificial-analysis · Bar chart · 25 points Estimated from the imageChat CompletionsOpenAI · Line chart · 9 points Estimated from the imageGPT-5.1 ranks #3 & #4 in Visionarena · Table chart · 45 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Bar chart · 34 points Estimated from the imageFactuality Leaderboardarena · Bar chart · 20 points Estimated from the imageTraining GPT-4o on 9x more compute would have yielded o1-high's performanceepoch-ai · Scatter plot · 6 points Estimated from the imageCode GPT-6 Astra (Max): Ranked #1arena · Bar chart · 24 points Estimated from the imageo1-high vastly outperforms GPT-4o on GPQA diamond and MATH level 5epoch-ai · Benchmark bar chart · 4 points Estimated from the imageMulti-Image Edit: Arena Ranked #1 & #2arena · Bar chart · 19 points Estimated from the imageText-to-Image Arena Ranks #2arena · Bar chart · 17 points Estimated from the imageInternal OpenAI Instructions following eval accuracy (hard subset)OpenAI · Bar chart · 8 points Estimated from the imageReal World Audio RecordingOpenAI · Bar chart · 4 points Estimated from the imageArtificial Analysis vs. Output Tokens Used in Artificial Analysis Intelligence Indexartificial-analysis · Scatter plot · 24 points Estimated from the imageOSWorld-VerifiedOpenAI · Bar chart · 3 points Estimated from the imageOpenAI o3 #2 GPT-4.1 & o4-mini in the top 10!arena · Table chart · 14 points Estimated from the imageFactual error rate on difficult prompts (lower is better)OpenAI · Line chart · 21 points Estimated from the imageGPT Image 2.5 Sunburst (max): Gap to Editing Action Frontierartificial-analysis · Table chart · 14 points Estimated from the imageContext Aware ASR BenchmarkOpenAI · Benchmark bar chart · 6 points Estimated from the imageOTIS Mock AIME accuracy for OpenAI's o3 and o4-mini compared to Gemini 2.5 Pro and leading modelsepoch-ai · Bar chart · 8 points Estimated from the imageArtificial Analysis Speech to Speech Indexartificial-analysis · Benchmark bar chart · 82 points Estimated from the imageOpenAI GPT-to-Image-1 ranks #1 in Text-to-Image Arena!arena · Table chart · 27 points Estimated from the imageReduction in AI scheming rates after deliberative alignmentOpenAI · Grouped bar chart · 4 points Estimated from the imageOpenAI o3-pro: pass@1 evaluation benchmarksOpenAI · Benchmark bar chart · 9 points Estimated from the imageo1 was a major algorithmic advanceepoch-ai · Grouped bar chart · 11 points Estimated from the imageOutput Tokens per Second; Higher is betterartificial-analysis · Bar chart · 18 points Estimated from the imageSEC-Bench Pro (Multi-Agent)OpenAI · Line chart · 17 points Estimated from the imageProfessional Capture-the-Flag ChallengesOpenAI · Scatter plot · 6 points Estimated from the imager²-Bench Telecom Benchmark Leaderboard: Resultsartificial-analysis · Bar chart · 32 points Estimated from the imageArtificial Analysis Cyber Index: Score vs. Cost per Taskartificial-analysis · Scatter plot · 19 points Estimated from the imageGPT-Live-1 Performance Bench v3 Tool callingOpenAI · Grouped bar chart · 3 points Estimated from the imageGeneration Time (Text to Image)artificial-analysis · Bar chart · 3 points Estimated from the imageCost per Hour of Input Audio (Big Bench Audio subset)artificial-analysis · Bar chart · 18 points Estimated from the imageArtificial Analysis Speech to Speech Index vs. time to first audio - Only models with all data available shown - Higher index and lower latency are betterartificial-analysis · Scatter plot · 13 points Estimated from the imageSpeech Reasoning (Big Bench Audio)artificial-analysis · Bar chart · 24 points Estimated from the imageArtificial Analysis Intelligence Index vs. Output Tokens per Intelligence Index Taskartificial-analysis · Scatter plot · 30 points Estimated from the imageOpenAI's Sora debuts on the Video Arena Leaderboard at #11arena · Table chart · 14 points Estimated from the imageSora 2 Pro tie ranks #1arena · Table chart · 12 points Estimated from the imageWarmth Scoreartificial-analysis · Bar chart · 5 points Estimated from the imageCode GPT-6 Sol: WebDev Ranked #4arena · Bar chart · 20 points Estimated from the imageArena Preference Elo vs. Task Success Rateartificial-analysis · Benchmark bar chart · 20 points Estimated from the imageSWE-Bench Pro (public)OpenAI · Line chart · 12 points Estimated from the imageMeta Llama 3 Instructartificial-analysis · Stacked bar chart · 12 points Estimated from the imageOutput tokens to run a set of aggregated benchmarkssemianalysis · Stacked bar chart · 32 points Estimated from the imageScore by Benchmarkartificial-analysis · Benchmark bar chart · 57 points Estimated from the imageInternal moderation eval (multi-policy accuracy)OpenAI · Bar chart · 6 points Estimated from the imageArena Rankings Updatearena · Table chart · 24 points Estimated from the imageGPT Image 2.5 Flare (max): Gap to Capability Frontierartificial-analysis · Table chart · 27 points Estimated from the imageOpen Weights: Frontier Indexartificial-analysis · Line chart · 30 points Estimated from the imageQuality vs Price Frontier (Image Editing)artificial-analysis · Scatter plot · 24 points Estimated from the imageArtificial Analysis Coding Agent Index v1.1artificial-analysis · Table chart · 16 points Estimated from the imageAverage OpenAI tool calls per model response across generationssemianalysis · Bar chart · 8 points Estimated from the imageIntelligence Index vs. Output Tokens per Intelligence Index Taskartificial-analysis · Scatter plot · 26 points Estimated from the imageArtificial Intelligence Indexartificial-analysis · Bar chart · 13 points Estimated from the imageHarvey LAB-AA v1.1: Hallucination Judge Model Comparisonartificial-analysis · Table chart · 66 points Estimated from the imageCompetition mathOpenAI · Benchmark bar chart · 10 points Estimated from the imageIntelligence vs. Total Parametersartificial-analysis · Scatter plot · 14 points Estimated from the imagetorchtitan gptosssemianalysis · Benchmark bar chart · 5 points Estimated from the imageBig Bench Audio: IntelligenceOpenAI · Bar chart · 2 points Estimated from the imageText-to-Image Arenaarena · Bar chart · 15 points Estimated from the imageGemini also ranks #1 on Text-to-Image Arenaarena · Table chart · 12 points Estimated from the imageQwen3.5-397B-A17B #17 Overall in Code Arenaarena · Table chart · 27 points Partly publisher's numbersInput Priceartificial-analysis · Bar chart · 21 points Estimated from the image#1 - Grok Imagine Videoarena · Table chart · 8 points Estimated from the imageArtificial Intelligence Analysis Intelligence Index: Non-Reasoning Modelsartificial-analysis · Bar chart · 16 points Estimated from the imageHarvey LAB-AA v1.1: Hallucination-Gated All-Pass Rate vs. Cost per Taskartificial-analysis · Scatter plot · 18 points Estimated from the imageTerminal Bench 2.0OpenAI · Bar chart · 2 points Estimated from the imageEnergy consumption per ChatGPT query is small compared to everyday electricity useepoch-ai · Bar chart · 7 points Estimated from the imageArtificial Analysis Cyber Index: Score vs. Cost per Taskartificial-analysis · Scatter plot · 22 points Estimated from the imagePareto Frontier for Agent Arenaarena · Scatter plot · 9 points Estimated from the imageOpenAI MRCRv2, 8 needlesOpenAI · Line chart · 10 points Estimated from the imageReplica 8.3Text-to-Image Arena Top 20arena · Bar chart · 25 points Estimated from the imageAgent 5.6: Sol, Terra & Lunaarena · Bar chart · 30 points Estimated from the imageReal World Audio RecordingOpenAI · Bar chart · 2 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 21 points Estimated from the imageConversational Dynamics (Full Duplex Bench v1.5 - Higher is better)artificial-analysis · Grouped bar chart · 16 points Estimated from the imageUsers long-running queries, run remotely, are disproportionately at the end of a user's last request of the dayOpenAI · Grouped bar chart · 15 points Estimated from the imageCritPt Benchmark Leaderboard: Token Usageartificial-analysis · Benchmark bar chart · 60 points Estimated from the imageAdvanced Cybersecurity Completion RateOpenAI · Bar chart · 4 points Estimated from the imageTime to First Audio (seconds) on Big Bench Audio dataset - Lower is betterartificial-analysis · Bar chart · 24 points Estimated from the imageImage Arena Quality ELOartificial-analysis · Bar chart · 17 points Estimated from the imageTau2-bench TelecomOpenAI · Bar chart · 5 points Estimated from the imageModel Performance Comparisonarena · Table chart · 79 points Estimated from the imageOSWorld 2.0, offline setOpenAI · Line chart · 15 points Estimated from the imageAgent Arenaarena · Table chart · 6 points Estimated from the imageModel Evaluation ScoresOpenAI · Benchmark bar chart · 40 points Estimated from the imageGPT-6Sol(Max) Pareto Frontier for Agent Arenaarena · Scatter plot · 11 points Estimated from the imageArtificial Analysis Intelligence Index v2.2artificial-analysis · Bar chart · 19 points Estimated from the imageGDP.pdfOpenAI · Line chart · 24 points Estimated from the imageArtificial Analysis Intelligence Index v4.1artificial-analysis · Table chart · 69 points Estimated from the imageCost per complete Hour of Input Audio (Big Bench Audio subset)artificial-analysis · Bar chart · 19 points Estimated from the imageTerminal-Bench 2.0OpenAI · Bar chart · 3 points Estimated from the imageTerminal-Bench 2.0OpenAI · Bar chart · 3 points Estimated from the imageAA-Briefcase v1.1: Scoreartificial-analysis · Grouped bar chart · 21 points Estimated from the imageCritPt Benchmark Leaderboard: Resultsartificial-analysis · Grouped bar chart · 18 points Estimated from the imageELO score in Artificial Analysis Image Arena (relative metric of image generation quality), Price: USD per 1000 image generationsartificial-analysis · Scatter plot · 20 points Estimated from the imageTau3 (Voice) IntelligenceOpenAI · Benchmark bar chart · 9 points Estimated from the imagePareto Frontier for Image-to-WebDevarena · Scatter plot · 10 points Estimated from the imageGrok Imagine Image 2.0: Quality vs Speed Frontier (Text to Image)artificial-analysis · Scatter plot · 19 points Estimated from the imageGoblins increased in GPT-5.4 especially for the Nerdy personalityOpenAI · Bar chart · 8 points Estimated from the imageCodeforces Competition codeOpenAI · Bar chart · 7 points Estimated from the imageAgents' Last ExamOpenAI · Line chart · 24 points Estimated from the imageArtificial Analysis Text to Video Leaderboardsartificial-analysis · Table chart · 12 points Estimated from the imageArtificial Analysis Cyber Index: Score by Benchmarkartificial-analysis · Benchmark bar chart · 51 points Estimated from the imageAgent Arenaarena · Bar chart · 25 points Estimated from the imageAudio MultiChallenge: Instruction FollowingOpenAI · Bar chart · 2 points Estimated from the imageSWE-Bench Verified (n=500)OpenAI · Line chart · 7 points Estimated from the imageGPT-5.2-chat-latest lands top 5 in Text Arenaarena · Table chart · 10 points Estimated from the imageText-to-Image-2.5 Arena: Ranked #1&#2arena · Bar chart · 18 points Estimated from the imageHumanity's Last ExamOpenAI · Bar chart · 7 points Estimated from the imageGPT-5 improves on o4-mini on SWE-bench Verified, but still scores below Claude 4.1epoch-ai · Bar chart · 6 points Estimated from the imageArtificial Analysis Intelligence Index vs. Cost per Intelligence Index Taskartificial-analysis · Scatter plot · 42 points Estimated from the imageFrontierMath private accuracy for OpenAI's o3 and o4-mini compared to leading modelsepoch-ai · Bar chart · 14 points Estimated from the imageComputer-use safety stress test (lower is better)OpenAI · Bar chart · 4 points Estimated from the imageSMB % of Agentic TokensOpenAI · Line chart · 8 points Estimated from the imageOverall model performanceOpenAI · Bar chart · 16 points

Anthropic 239

Estimated from the imageClaude 3.5 Sonnet benchmarksAnthropic · Table chart · 34 points Estimated from the imageSoftware engineeringAnthropic · Bar chart · 2 points Estimated from the imageNovel problem-solving by costAnthropic · Scatter plot · 3 points Estimated from the imageReplica 7.4Code Arena: WebDevarena · Bar chart · 15 points Estimated from the imageModel benchmarksAnthropic · Table chart · 31 points Estimated from the imageReplica 6.0Opus 4.8Anthropic · Table chart · 28 points Estimated from the imageFrontend Code Kimi-K3: Ranked #1arena · Bar chart · 23 points Estimated from the imageClaude 4 benchmarksAnthropic · Table chart · 65 points Estimated from the imageReplica 6.0Artificial Analysis Intelligence Index v4.1artificial-analysis · Grouped bar chart · 30 points Estimated from the imageReplica 7.1Artificial Analysis Intelligence Index v4.1artificial-analysis · Bar chart · 44 points Estimated from the imageClaude Mythos 5 and Fable 5Anthropic · Table chart · 67 points Estimated from the imageOpus 5Anthropic · Table chart · 51 points Estimated from the imageReplica 2.4Firefox Security Vulnerabilities by MonthAnthropic · Stacked bar chart · 56 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Bar chart · 35 points Estimated from the imageReplica 7.7Long context retrievalAnthropic · Line chart · 16 points Estimated from the imageReplica 6.6Agentic business Automation workflows by effortAnthropic · Line chart · 36 points Estimated from the imageAgentic codingAnthropic · Line chart · 31 points Estimated from the imageReplica 7.8Model Performance ComparisonAnthropic · Table chart · 46 points Estimated from the imageReplica 7.7Agentic computer useAnthropic · Scatter plot · 18 points Estimated from the imageReplica 8.3Agentic computer use performance by effort levelAnthropic · Line chart · 19 points Estimated from the imageReplica 9.3SoftwareAnthropic · Bar chart · 6 points Estimated from the imageAgentic terminalAnthropic · Line chart · 21 points Estimated from the imageCursorBench3.2.0 Agentic codingAnthropic · Line chart · 10 points Estimated from the imageReplica 7.4SoftwareAnthropic · Bar chart · 6 points Estimated from the imageAgentic codingAnthropic · Line chart · 21 points Estimated from the imageSonnet 5Anthropic · Table chart · 18 points Partly publisher's numbersClaude Haiku 5.5Anthropic · Table chart · 42 points Estimated from the imageHumanity's Last Exam Multidisciplinary reasoningAnthropic · Line chart · 36 points Estimated from the imageClaude 3.7 Sonnet benchmarksAnthropic · Table chart · 56 points Estimated from the imageReplica 8.0Business workflowsAnthropic · Line chart · 21 points Estimated from the imageWhere a researcher went wrong, could Claude have done better?Anthropic · Benchmark bar chart · 20 points Partly publisher's numbersClaude Fable 5.1Anthropic · Table chart · 53 points Estimated from the imageModel Benchmark ComparisonAnthropic · Table chart · 77 points Estimated from the imageIndexed cost of Fable usageAnthropic · Benchmark bar chart · 12 points Estimated from the imageReplica 2.7MultidisciplinaryAnthropic · Line chart · 30 points Estimated from the imageReplica 4.0Real-world knowledgeAnthropic · Line chart · 25 points Estimated from the imageOpus 4.7Anthropic · Table chart · 64 points Estimated from the imageAI usage by job typeAnthropic · Grouped bar chart · 42 points Estimated from the imageReplica 8.0Sonnet 4.6Anthropic · Table chart · 83 points Estimated from the imageFinding vulnerabilities and exploits in open-source codeAnthropic · Grouped bar chart · 12 points Partly publisher's numbersClaude Opus 5.5Anthropic · Table chart · 39 points Estimated from the imageComputer useAnthropic · Line chart · 10 points Estimated from the imageTerminal-Bench-Science 0.1 Agentic scientific researchAnthropic · Line chart · 10 points Estimated from the imageAgentic codingAnthropic · Line chart · 22 points Estimated from the imageTerminal-Bench 4.0 Agentic terminal codingAnthropic · Line chart · 15 points Estimated from the imageSoftwareAnthropic · Bar chart · 6 points Estimated from the imageAgentic terminalAnthropic · Line chart · 27 points Estimated from the imageMisaligned behaviorAnthropic · Bar chart · 4 points Estimated from the imageKnowledgework by effort levelAnthropic · Line chart · 20 points Estimated from the imageGPT-6.1 Sol and Claude Opus 5.5: API value by plansemianalysis · Grouped bar chart · 6 points Estimated from the imageAgent Arenaarena · Bar chart · 14 points Estimated from the imagePareto Frontier for Text Arenaarena · Scatter plot · 9 points Estimated from the imageAgentic coding runs on cachesemianalysis · Table chart · 18 points Estimated from the imageAA-Briefcase Presentation Elo vs. Rubric Score (%)artificial-analysis · Scatter plot · 38 points Estimated from the imageSycophancy by model (levels 1-2)Anthropic · Grouped bar chart · 8 points Estimated from the imageIntelligence Analysis Index vs. Cost per Taskartificial-analysis · Scatter plot · 25 points Estimated from the imageArena Score Top15arena · Bar chart · 13 points Estimated from the imageGDPval-AA Leaderboardartificial-analysis · Bar chart · 27 points Estimated from the imageMultilingual Index (English)artificial-analysis · Bar chart · 12 points Estimated from the imageCritPt Benchmark Developed by Argonne and UIUC: Scoreartificial-analysis · Bar chart · 35 points Estimated from the imageOutput Tokens per Secondartificial-analysis · Bar chart · 17 points Estimated from the imageCost per Token type Intelligence Index Taskartificial-analysis · Table chart · 30 points Estimated from the imageHow Claude Models Writearena · Grouped bar chart · 24 points Estimated from the imageAutomationBenchopenai · Line chart · 27 points Estimated from the imageClaude Opus 5.5 leads on software engineering, GPT-6 Astra on mathepoch-ai · Table chart · 11 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 25 points Estimated from the imageElo rating-AA for performance on real-world work tasks - Anchored to a human baseline of 1,000 - Higher is betterartificial-analysis · Bar chart · 36 points Estimated from the imageGDPpdf: All-passartificial-analysis · Grouped bar chart · 26 points Estimated from the imageExpert GPT-5.5: Top #5arena · Bar chart · 17 points Estimated from the imageGDPval scores for agentic performance on real-world work tasks using web and shell access via Stirrup, an open-source harness developed by Artificial Analysisartificial-analysis · Bar chart · 33 points Estimated from the imageClaude 3.5 Sonnet (new) benchmarksAnthropic · Table chart · 26 points Estimated from the imageClaude Opus 5.5 (High) & Pareto in Agent Arenaarena · Bar chart · 15 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Grouped bar chart · 11 points Estimated from the imageHow long does Claude Code work before stopping?Anthropic · Line chart · 69 points Estimated from the imageGPQA Diamond accuracy for Claude Sonnet 4 and Claude Opus 4 compared to other modelsepoch-ai · Bar chart · 9 points Estimated from the imageAA-Share of tasks Agent solved on all 5 of pass^5 Leaderboardartificial-analysis · Bar chart · 36 points Estimated from the imageTerminal-Bench v4.0: Scoreartificial-analysis · Benchmark bar chart · 36 points Estimated from the imageArtificial Analysis Intelligence Index vs. Output Tokens per Intelligence Index Taskartificial-analysis · Scatter plot · 26 points Estimated from the imageSoftware engineeringAnthropic · Benchmark bar chart · 10 points Estimated from the imageAA-Briefcase Rubric Score (%)artificial-analysis · Grouped bar chart · 39 points Estimated from the imageCode Arena: Frontendarena · Bar chart · 25 points Estimated from the imageCoding, Hard Prompts, Writing & Longer Query improved significantly over 3.7arena · Table chart · 119 points Estimated from the imageHow Claude Models Writearena · Grouped bar chart · 28 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 39 points Estimated from the imageTotal Coding Market ARRsemianalysis · Line chart · 132 points Estimated from the imageIntelligence vs. Time per Intelligence Index Taskartificial-analysis · Scatter plot · 70 points Estimated from the imageHarvey LAB-AA: All-pass rate vs. Average cost per Task (USD)artificial-analysis · Scatter plot · 29 points Estimated from the imageKimi K3 is the highest-ECI open model to date, CIs are 90% confidence intervalsepoch-ai · Benchmark bar chart · 11 points Estimated from the imageMean Turns per Taskartificial-analysis · Bar chart · 36 points Estimated from the imageIntelligence Analysis vs. Output Tokensartificial-analysis · Scatter plot · 31 points Estimated from the imageArtificial Analysis Coding Agent Indexartificial-analysis · Grouped bar chart · 13 points Estimated from the imageGPQA Diamond (Scientific Reasoning)artificial-analysis · Bar chart · 15 points Estimated from the imageEstimating Claude's productivity impactsAnthropic · Grouped bar chart · 8 points Estimated from the imageAgent Arena: Codearena · Bar chart · 16 points Estimated from the imageArtificial Analysis Intelligence Index v4.1artificial-analysis · Benchmark bar chart · 61 points Estimated from the imageArtificial Analysis Intelligence Index incorporates 7 evaluationsartificial-analysis · Bar chart · 19 points Estimated from the imageAA-Omniscence Index: Scoreartificial-analysis · Benchmark bar chart · 79 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Bar chart · 31 points Estimated from the imageHarvey Share of LAB tasks-AA: All-pass criterion Rateartificial-analysis · Bar chart · 37 points Estimated from the imageMultilingual Index (French)artificial-analysis · Bar chart · 12 points Estimated from the imageWhat a dollar of ChatGPT and Claude buys todaysemianalysis · Scatter plot · 11 points Estimated from the imageHow Claude Models Writearena · Grouped bar chart · 29 points Estimated from the imageSWE-bench Verified accuracy for Claude Sonnet 4 and Claude Opus 4 compared to other modelsepoch-ai · Bar chart · 18 points Estimated from the imageNet Production Models PerformanceAnthropic · Benchmark bar chart · 25 points Estimated from the imageSWE-Atlas@1 on SWE-Atlas-QnA - Higher is betterartificial-analysis · Bar chart · 19 points Estimated from the imageClaude 3.7 Sonnet gain was a major algorithmic advanceepoch-ai · Benchmark bar chart · 12 points Estimated from the imageAgent GPT-5.5 (xHigh) Leaderboard: Ranked #2arena · Benchmark bar chart · 20 points Estimated from the imageHow the work changed over seven monthsAnthropic · Line chart · 63 points Estimated from the imageCode Arena: WebDevarena · Bar chart · 15 points Estimated from the imageTime per Taskartificial-analysis · Grouped bar chart · 38 points Estimated from the imageCritPt Benchmark Leaderboard: Resultsartificial-analysis · Grouped bar chart · 32 points Estimated from the imageWhat kinds of failures are flagged?arena · Table chart · 115 points Estimated from the imageArtificial Analysis Intelligence Index v2.1artificial-analysis · Table chart · 36 points Estimated from the imageClaude Opus 4.7 ranks #1 in Code Arenaarena · Table chart · 20 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Bar chart · 23 points Estimated from the imageLiveCodeBench (Coding): Non-Reasoning Modelsartificial-analysis · Bar chart · 12 points Estimated from the imageSuccessAnthropic · Grouped bar chart · 31 points Estimated from the imageTerminal-Bench-Science 0.1 score vs. average Cost per Task (USD) - Lower is betterartificial-analysis · Scatter plot · 134 points Estimated from the imageEnterpriseOps-Gym-AA score vs. average cost per task (Cost USD) - Lower is betterartificial-analysis · Scatter plot · 28 points Estimated from the imageAA-Briefcase Elo vs. Cost per Taskartificial-analysis · Scatter plot · 28 points Estimated from the imageArtificial Analysis Coding Agent Index v1.4artificial-analysis · Bar chart · 21 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Bar chart · 36 points Estimated from the imageOutput Tokens used Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 36 points Estimated from the imageArtificial Analysis Intelligence Index v4.1artificial-analysis · Bar chart · 42 points Estimated from the imageClaude Opus 4.7 Thinking ranks #1 in Vision Arenaarena · Table chart · 20 points Estimated from the imageSWE-bench Verified accuracy for Claude Sonnet 4 and Claude Opus 4 compared to previous Claude modelsepoch-ai · Bar chart · 10 points Estimated from the imageClaude Opus 4 tied #1 in WebDev Arenaarena · Table chart · 33 points Estimated from the imageSoftware engineeringAnthropic · Bar chart · 6 points Estimated from the imageFrontierCode 1.1: Score vs. Cost per Taskartificial-analysis · Scatter plot · 29 points Estimated from the imageClaude models playing Pokémon*Anthropic · Line chart · 44 points Estimated from the imageAnthropic in the Text Arenaarena · Table chart · 60 points Estimated from the imageCode Arena: Frontendarena · Bar chart · 17 points Estimated from the imageAverage cost per Taskartificial-analysis · Bar chart · 19 points Estimated from the imageFraction of examples with faithful CoTAnthropic · Benchmark bar chart · 4 points Estimated from the imageHow Claude Models Writearena · Grouped bar chart · 24 points Estimated from the imageClaude Opus 4.1 ranks #2 on WebDev (tied with Gemini 2.5 Pro)arena · Table chart · 11 points Estimated from the imageMythos Preview outpaced the trend in cyber frontier models (best scores at release)epoch-ai · Line chart · 28 points Estimated from the imageAA-AnalystAgent:pass^5 vs Cost per Taskartificial-analysis · Scatter plot · 30 points Estimated from the imageArena Opus 4.7 vs Opus 4.6 Text Category Rankingsarena · Line chart · 28 points Estimated from the imageImage-to-WebDev Arena Top 25: Overallarena · Bar chart · 25 points Estimated from the imageIntelligence Index vs. Cost per Intelligence Index Taskartificial-analysis · Scatter plot · 12 points Estimated from the imageClaude Opus 4.7 ranks #1 in Document Arenaarena · Table chart · 20 points Estimated from the imageGPT-5 is #1 on WebDev Arenaarena · Table chart · 8 points Estimated from the imageAgent Arena Leaderboardarena · Bar chart · 25 points Estimated from the imageHow Claude Models Writearena · Grouped bar chart · 29 points Estimated from the imageClaude Code interruption rate by experienceAnthropic · Line chart · 16 points Estimated from the imageMultilingual Index (Spanish)artificial-analysis · Bar chart · 13 points Estimated from the imageMMLU (Quality - Claude 3 comparison)artificial-analysis · Bar chart · 10 points Estimated from the imageTask Complexity, Max. Consecutive Tool Calls, and Number of Human TurnsAnthropic · Benchmark bar chart · 6 points Estimated from the imageWebDev Arena Leaderboardarena · Table chart · 16 points Estimated from the imageHarvey LAB-AA all-pass: All-pass Rate vs. Output Tokens per Taskartificial-analysis · Scatter plot · 29 points Estimated from the imageWebDev Arena Leaderboardarena · Table chart · 24 points Estimated from the imageIntelligence vs. Cost to Run Artificial Analysis Intelligence Indexartificial-analysis · Scatter plot · 17 points Estimated from the imageArtificial Analysis Coding Index (Non-Reasoning Models)artificial-analysis · Bar chart · 12 points Estimated from the imageScore by Benchmarkartificial-analysis · Benchmark bar chart · 66 points Estimated from the imageTerminal pass@1 on Terminal-Bench v2artificial-analysis · Bar chart · 21 points Estimated from the imageArtificial Analysis Intelligence Index v4.1artificial-analysis · Grouped bar chart · 27 points Estimated from the imageMultidisciplinaryAnthropic · Line chart · 10 points Estimated from the imageClaude Sonnet 4.5 #4 for WebDev Arenaarena · Table chart · 36 points Estimated from the imageSciCode: Scoreartificial-analysis · Benchmark bar chart · 64 points Estimated from the imagePlan usage limitsAnthropic · Bar chart · 3 points Estimated from the imageClaude Opus 5.5 (High) Pareto Frontier for Agent Arenaarena · Line chart · 10 points Estimated from the imageMMLU vs. Priceartificial-analysis · Scatter plot · 10 points Estimated from the imageHow Claude Models Writearena · Grouped bar chart · 24 points Estimated from the imageToken Usage, cache, and Output tokens per Taskartificial-analysis · Stacked bar chart · 75 points Estimated from the imageArtificial Analysis Coding Indexartificial-analysis · Bar chart · 20 points Estimated from the imageOutput Tokens per Intelligence Index Taskartificial-analysis · Scatter plot · 40 points Estimated from the imageHow Claude Models Writearena · Table chart · 27 points Estimated from the imageDeepSWE Benchmark Scoreartificial-analysis · Bar chart · 15 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 20 points Estimated from the imageDeepSeek V3.1 excels at Math, Creative Writing and Longer Query (Top 3)arena · Table chart · 135 points Estimated from the imageIterative improvement on a frontier-scale modelAnthropic · Line chart · 59 points Estimated from the imageOutputartificial-analysis · Bar chart · 41 points Estimated from the imageAA-Briefcase Elo vs. Cost per Taskartificial-analysis · Scatter plot · 63 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Grouped bar chart · 35 points Estimated from the imageAutomationBench-AA: Scoreartificial-analysis · Grouped bar chart · 35 points Estimated from the imageArtificial Analysis Coding Agent Indexartificial-analysis · Bar chart · 10 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Bar chart · 40 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 20 points Estimated from the imageCode GPT-6 Astra: WebDev Win Rates in Battlesarena · Grouped bar chart · 15 points Estimated from the imageArtificial Analysis Legal Index - Weighted average wall clock time per Taskartificial-analysis · Scatter plot · 22 points Estimated from the imageIntelligence Analysis vs. Cost per Intelligence Index Taskartificial-analysis · Scatter plot · 80 points Estimated from the imageText #5 Labs: Best Models Category Rankingsarena · Table chart · 40 points Estimated from the imageOptimize throughput with prompt cachingAnthropic · Stacked bar chart · 4 points Estimated from the imageCode Arena Top 15: Agentic Codingarena · Bar chart · 14 points Estimated from the imageOutput Tokens per Intelligence Index Taskartificial-analysis · Benchmark bar chart · 70 points Estimated from the imageClaude Code+ Commits Over Timesemianalysis · Line chart · 12 points Estimated from the imageText Arenaarena · Bar chart · 17 points Estimated from the imageTest cases passed for Opus 4.6 reimplementing pkl in Rustepoch-ai · Line chart · 105 points Estimated from the imageArtificial Analysis Intelligence Index v4.3artificial-analysis · Grouped bar chart · 28 points Estimated from the imageLatency vs. Throughput: Provider Chunk & Model combinationsartificial-analysis · Scatter plot · 8 points Estimated from the imageModel Benchmark Comparisonopenai · Table chart · 43 points Estimated from the imageHow Claude Models Write Language Markersarena · Grouped bar chart · 28 points Estimated from the imageNewer models score higher for same token budgetepoch-ai · Line chart · 12 points Estimated from the imageFrontier Analysis Language Intelligence Index v4.3.2 Intelligence By Country, Over Timeartificial-analysis · Line chart · 57 points Estimated from the imageTask duration where we predict 50% chance of AI successmetr · Bar chart · 6 points Estimated from the imageClaude 3.5 Haiku benchmarksAnthropic · Table chart · 26 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Bar chart · 34 points Estimated from the imageArtificial Analysis Indexesartificial-analysis · Benchmark bar chart · 80 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Bar chart · 33 points Estimated from the imageCode Arena Top 15: Agentic Codingarena · Bar chart · 14 points Estimated from the imageClaude tool calls per model responsesemianalysis · Bar chart · 5 points Estimated from the imageGDPval-AA v2 Leaderboardartificial-analysis · Bar chart · 25 points Estimated from the imageArtificial Analysis Intelligence Index v4.1artificial-analysis · Bar chart · 26 points Estimated from the imageTerminal-Bench-Science 0.1 score vs. average Cost per Task (USD) - Lower is betterartificial-analysis · Scatter plot · 12 points Estimated from the imageUsage Share Across Occupational CategoriesAnthropic · Grouped bar chart · 15 points Estimated from the imageHarvey LAB-AA: All-pass Rate vs. Time per Taskartificial-analysis · Scatter plot · 30 points Estimated from the imageHow Claude Models Write Language Markersarena · Grouped bar chart · 24 points Estimated from the imageReplica 3.0Artificial Analysis Intelligence Indexartificial-analysis · Grouped bar chart · 18 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Bar chart · 29 points Estimated from the imageAA-Briefcase v1.1artificial-analysis · Grouped bar chart · 53 points Estimated from the imageMMLU (Quality benchmark comparison)artificial-analysis · Bar chart · 11 points Estimated from the imageOutput Tokens per Intelligence Index Taskartificial-analysis · Table chart · 22 points Estimated from the imageReal-world knowledgeAnthropic · Line chart · 16 points Estimated from the imageFraction of examples with faithful CoTAnthropic · Grouped bar chart · 5 points Estimated from the imageClarification and interruption rates by goal complexityAnthropic · Grouped bar chart · 8 points Estimated from the imageAgent Arena Claude Sonnet 5arena · Table chart · 6 points Estimated from the imageImage-to-WebDev Arena Ranked #2arena · Bar chart · 15 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Benchmark bar chart · 88 points Estimated from the imageMultilingual Index (German)artificial-analysis · Bar chart · 12 points Estimated from the imagePareto Frontier for Agent Arenaarena · Scatter plot · 13 points Estimated from the imagePricing per 1M Tokensartificial-analysis · Grouped bar chart · 24 points Estimated from the imageCode Arena: WebDev GPT-6 Astra vs Claude Fable 5.1arena · Table chart · 16 points Estimated from the imageText Arena Claude Sonnet 5 vs 4.6 Category Rankingsarena · Table chart · 30 points Estimated from the imageHarvey LAB-AA Benchmark Leaderboard: All-pass Rate vs. Average Turns per Taskartificial-analysis · Scatter plot · 40 points Estimated from the imageTerminal-Bench Hard Artificial Analysis Leaderboard: Resultsartificial-analysis · Bar chart · 31 points Estimated from the imageAI performance solving issues from 12 open-source Python repositoriesepoch-ai · Scatter plot · 22 points Estimated from the imageClaude Opus 4.7 scores near the frontierepoch-ai · Bar chart · 5 points Estimated from the imageHow Claude models differAnthropic · Grouped bar chart · 12 points Estimated from the imageTerminal-Bench 4.0: Scoreartificial-analysis · Benchmark bar chart · 36 points Estimated from the imageGemini 2.5 Pro also jumps to #2 on WebDev Arena!arena · Table chart · 30 points Estimated from the imagePareto Frontier for Code Arenaarena · Line chart · 7 points Estimated from the imageCode Arena: WebDevarena · Table chart · 14 points Estimated from the imageOutput Tokens per Intelligence Index Taskartificial-analysis · Scatter plot · 24 points Estimated from the imageClaude 3.7 Sonnet Thinking shows strength across a variety of categoriesarena · Table chart · 150 points Estimated from the imageMisaligned cyberattacks in pursuit of reward...Anthropic · Grouped bar chart · 8 points Estimated from the imageMisaligned actions in pursuit of rewardAnthropic · Grouped bar chart · 8 points Estimated from the imageSimulation Inspired by Hugging Face IncidentAnthropic · Grouped bar chart · 31 points Estimated from the imageUS GDP (trillions of dollars a year, in 2025 prices)Anthropic · Line chart · 30 points

Google 179

Estimated from the imageReplica 8.3Gemini 3.8 Flash CyberGoogle · Bar chart · 3 points Estimated from the imageGemini 2.5 Flash Native AudioGoogle · Benchmark bar chart · 7 points Estimated from the imageCWE-bench v1 leaderboardGoogle · Bar chart · 11 points Estimated from the imageGemini 3 Deep ThinkGoogle · Benchmark bar chart · 18 points Estimated from the imageReplica 8.2Speech Reasoning: Big Bench Audioartificial-analysis · Bar chart · 14 points Estimated from the imageDiscovering securityGoogle · Benchmark bar chart · 4 points Estimated from the imageEva Bench - Experience to accuracyGoogle · Scatter plot · 13 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Bar chart · 27 points Estimated from the imageReplica 9.2Gemini 4 Argon Benchmark ComparisonGoogle · Table chart · 88 points Estimated from the imageReplica 6.1Code Gemini 4 Argon: WebDev Ranked #8arena · Bar chart · 39 points Estimated from the imageWebDev Arena Leaderboard 05-06-25 02:00 PTGoogle · Table chart · 48 points Estimated from the imageText Arena Gemini 4 Argon: Ranked #1arena · Bar chart · 25 points Estimated from the imageGray Swan IPIGoogle · Stacked bar chart · 42 points Estimated from the imageReplica 7.8Artificial Analysis Intelligence Index vs. Cost to Run Artificial Analysis Intelligence Indexartificial-analysis · Scatter plot · 34 points Estimated from the imageArena Score vs Cost per 1M Output TokensGoogle · Scatter plot · 34 points Estimated from the imageMassive Audio Embedding BenchmarkGoogle · Scatter plot · 9 points Estimated from the imageReplica 7.6Model Performance VS SizeGoogle · Scatter plot · 10 points Estimated from the imageReplica 7.6Generalization BenchmarkGoogle · Benchmark bar chart · 9 points Estimated from the imageText to ImageGoogle · Benchmark bar chart · 18 points Estimated from the imageSpeed & Cost EfficiencyGoogle · Grouped bar chart · 29 points Estimated from the imageUnlock token efficiency and accuracy gainsGoogle · Benchmark bar chart · 12 points Estimated from the imageHumanity's Last ExamGoogle · Bar chart · 4 points Estimated from the imageCodeforcesGoogle · Bar chart · 3 points Estimated from the imageSpeech to SpeechGoogle · Bar chart · 6 points Estimated from the imageGemini 3.6 FlashGoogle · Benchmark bar chart · 12 points Estimated from the imageEnterprise workflow automationGoogle · Bar chart · 4 points Estimated from the imageFLEURS (top locales*)Google · Bar chart · 5 points Estimated from the imageGemini 2.5 Pro Deep ThinkGoogle · Benchmark bar chart · 12 points Estimated from the imageReplica 7.0Speech to Speech IndexGoogle · Bar chart · 6 points Estimated from the imageTime to solve (minutes)Google · Grouped bar chart · 21 points Estimated from the imageHumanity's Last ExamGoogle · Bar chart · 4 points Estimated from the imageLong-horizon software engineeringGoogle · Bar chart · 5 points Estimated from the imageProduction FrontierCode 1.1 MainGoogle · Bar chart · 4 points Estimated from the imageAgentic Performance (t-Voice)Google · Bar chart · 6 points Estimated from the imageASIMOV - Safety Instruction FollowingGoogle · Grouped bar chart · 9 points Estimated from the imageGemini 3.5 Flash-Lite Compared to 3 FlashGoogle · Benchmark bar chart · 4 points Estimated from the imageMMLU Performance vs. Model Sizes (Mix-n-Match Configurations)Google · Scatter plot · 16 points Estimated from the imageτ³-Banking LeaderboardGoogle · Bar chart · 5 points Estimated from the imageGemini 3.7 FlashGoogle · Bar chart · 5 points Estimated from the imageMassive Image Embedding Benchmark (Lite)Google · Scatter plot · 9 points Estimated from the imageAgentic PerformanceGoogle · Bar chart · 6 points Estimated from the imageMMMU-ProGoogle · Bar chart · 4 points Estimated from the imageFLEURS (top locales*)Google · Bar chart · 6 points Estimated from the imageMassive Text Embedding Benchmark (Code)Google · Scatter plot · 9 points Estimated from the imageGemini 3 Deep ThinkGoogle · Bar chart · 4 points Estimated from the imageModel Capability ComparisonGoogle · Table chart · 33 points Estimated from the image3.7 Flash is the Agentic frontier video understanding on accuracy and costGoogle · Scatter plot · 6 points Estimated from the imageGemini 3.5 Flash-LiteGoogle · Benchmark bar chart · 4 points Estimated from the imageInstruction Following BenchmarkGoogle · Benchmark bar chart · 6 points Estimated from the imageArtificial Intelligence Analysis incorporates Intelligence Indexartificial-analysis · Bar chart · 22 points Estimated from the imageITBench-AA: Average precision at full recall vs. Cost per Taskartificial-analysis · Scatter plot · 38 points Estimated from the imageNano Banana 2.1: Gap to Capability Frontier (AA-Image-Editing v2.0)artificial-analysis · Table chart · 21 points Estimated from the imagePriceartificial-analysis · Bar chart · 29 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Bar chart · 35 points Estimated from the imageCost (USD) to run all evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 29 points Estimated from the imageELO scores from blind preference votes in our Image Arena. See the full leaderboard here.artificial-analysis · Bar chart · 17 points Estimated from the imageGemini 4 Argon Pareto Frontier for Text Arenaarena · Scatter plot · 10 points Estimated from the imageGemini 4 Argon Pareto Frontier for Text Arenaarena · Scatter plot · 10 points Estimated from the imageArena Code Agentic WebDev Leaderboardarena · Table chart · 25 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Grouped bar chart · 26 points Estimated from the imageOutput Tokens per Secondartificial-analysis · Bar chart · 30 points Estimated from the imageSeedream 3.0 ties #4 in Text-to-Image Arena!arena · Table chart · 27 points Estimated from the imageCode Arena: Frontendarena · Bar chart · 17 points Estimated from the imageArtificial Analysis Video Arena Leaderboard (Image to Video)artificial-analysis · Table chart · 18 points Estimated from the imageChatbot Arena Overview (Task)arena · Table chart · 117 points Estimated from the imagePareto Frontier for Agent Arenaarena · Line chart · 1 points Estimated from the imageOmniscience Accuracy vs. Omniscience Indexartificial-analysis · Scatter plot · 33 points Estimated from the imageMMMU-Pro Benchmark Leaderboard: Resultsartificial-analysis · Bar chart · 24 points Estimated from the imageArtificial Analysis Intelligence Index vs. Cost per Taskartificial-analysis · Scatter plot · 34 points Estimated from the imageNano Banana 2.1: Gap to Capability Frontier (AA-Image-T2I v2.0)artificial-analysis · Table chart · 27 points Estimated from the imageSingle Image Edit Arenaarena · Bar chart · 14 points Estimated from the imageIntelligence vs. Priceartificial-analysis · Scatter plot · 12 points Estimated from the imageGPQA Diamond accuracy for Mistral Small 3 compared to other modelsepoch-ai · Bar chart · 5 points Estimated from the imageCharacters Per secondartificial-analysis · Bar chart · 44 points Estimated from the imageText Arenaarena · Bar chart · 31 points Estimated from the imageIntelligence Parameters vs. Model Parameters: Intelligence Small, Open Weights (Version 2, released Feb '25)artificial-analysis · Scatter plot · 11 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Bar chart · 100 points Estimated from the imageMMMU-Pro Benchmark Leaderboard: Resultsartificial-analysis · Bar chart · 19 points Estimated from the imageIntelligence vs. Compute Proxyartificial-analysis · Scatter plot · 17 points Estimated from the imageGPQA Diamond accuracy for OpenAI's o3 and o4-mini compared to leading modelsepoch-ai · Bar chart · 7 points Estimated from the imagePronunciation Robustness by Categoryartificial-analysis · Grouped bar chart · 64 points Estimated from the imageFigure 2: Average Win Rate Against All Other Models (Assuming Uniform Sampling and No Ties)arena · Bar chart · 8 points Estimated from the imageImage Nano Banana 2.1: Ranked #6arena · Bar chart · 16 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Grouped bar chart · 26 points Estimated from the imageSpeech reasoning: based on the Artificial Analysis Big Bench Audio dataset - Higher is betterartificial-analysis · Bar chart · 33 points Estimated from the imagePareto Frontier for Agent Arenaarena · Scatter plot · 9 points Estimated from the imageAA-WER Streaming Index (First Partial Transcription) vs. Time to First Partial Transcription After Speech Endartificial-analysis · Scatter plot · 28 points Estimated from the imageOutput Reasoning Tokens Used to Run Artificial Analysis Intelligence Indexartificial-analysis · Grouped bar chart · 38 points Estimated from the imageNano Banana 2.1: Quality vs Speed Frontier (AA-Image-T2I v2.0)artificial-analysis · Scatter plot · 7 points Estimated from the imageNano Banana 2.1: Quality vs Price Frontier (AA-Image-T2I v2.0)artificial-analysis · Scatter plot · 12 points Estimated from the imageOutput Tokens used Run all evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 76 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Grouped bar chart · 44 points Estimated from the imageArtificial Analysisartificial-analysis · Benchmark bar chart · 40 points Estimated from the imageGDPval-AA v2 Leaderboardartificial-analysis · Grouped bar chart · 42 points Estimated from the imageSpeech Reasoning: Based on the Artificial Analysis Big Bench Audio dataset; Higher is betterartificial-analysis · Bar chart · 13 points Estimated from the imageDeepSWE v1.1semianalysis · Scatter plot · 10 points Estimated from the imageArtificial Analysis Intelligence Index incorporates 7 evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, AIME, MATH-500artificial-analysis · Bar chart · 19 points Estimated from the imageCost per Intelligence Index Taskartificial-analysis · Bar chart · 45 points Estimated from the imageArtificial Analysis Intelligence Index v4.1artificial-analysis · Grouped bar chart · 40 points Estimated from the imageOutput Tokens per Secondartificial-analysis · Bar chart · 31 points Estimated from the imageText-to-Image Arena 2.1: Ranked #5arena · Bar chart · 15 points Estimated from the imageTime per Intelligence Index Taskartificial-analysis · Bar chart · 25 points Estimated from the imageImage-to-Video Arena 2025arena · Table chart · 19 points Estimated from the imageArtificial Analysis Omniscience Accuracyartificial-analysis · Bar chart · 35 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Bar chart · 35 points Estimated from the imageText Arenaarena · Bar chart · 26 points Estimated from the imageText-to-Video Arena 2025arena · Table chart · 40 points Estimated from the imageCost per Hour of Input Audio Big Bench (Big Bench Audio subset)artificial-analysis · Bar chart · 20 points Estimated from the imageGDPval-AA Leaderboardartificial-analysis · Bar chart · 27 points Estimated from the imageIntelligence vs. Total Parametersartificial-analysis · Scatter plot · 49 points Estimated from the imageOutput Tokensartificial-analysis · Benchmark bar chart · 78 points Estimated from the imageProvider Elo: Arena Preferenceartificial-analysis · Bar chart · 38 points Estimated from the imageIntelligence vs. Cost to Run Artificial Analysis Intelligence Indexartificial-analysis · Scatter plot · 30 points Estimated from the imageGPT-5 matches the performance of other models on GPQA Diamondepoch-ai · Bar chart · 8 points Estimated from the imageArtificial Analysis Intelligence Index incorporates 7 evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, AIME, MATH-500artificial-analysis · Bar chart · 22 points Estimated from the imageOutput Tokens per Secondartificial-analysis · Bar chart · 22 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Bar chart · 17 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Bar chart · 37 points Estimated from the imageAgentic conversation performance (τ-Voice) Average conversation Audio Durationartificial-analysis · Bar chart · 12 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Benchmark bar chart · 231 points Estimated from the imageVideo Arenaarena · Bar chart · 14 points Estimated from the imageRank from human voice preference votes on the Artificial Analysis Speech Arena (Controlled voice). Rank is calculated independently for each language. Lower rank is better.artificial-analysis · Table chart · 110 points Estimated from the imageArtificial Analysis Intelligence Index v4.1artificial-analysis · Bar chart · 39 points Estimated from the imageFigure 1: Confidence Intervals on Model Imagen Strength (via Bootstrapping)arena · Scatter plot · 8 points Estimated from the imageOutput Tokens Used to Run Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 65 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 25 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Bar chart · 24 points Estimated from the imagePareto Frontierarena · Scatter plot · 8 points Estimated from the imageAgent Arenaarena · Bar chart · 25 points Estimated from the imageIntelligence Index vs. Total Parametersartificial-analysis · Scatter plot · 15 points Estimated from the imageOutput speed: o200k_base Gemma 4 31B (Reasoning)epoch-ai · Bar chart · 13 points Estimated from the imageVeo-3.1-1080p ranks #1 & 2 in Video Arenaarena · Table chart · 30 points Estimated from the imageIntelligence vs. Output Speedartificial-analysis · Scatter plot · 33 points Estimated from the imageKey Players in the AI Value Chainartificial-analysis · Table chart · 96 points Estimated from the imageAgent Arena Gemini 3.8 Flash vs Gemini 3.7 Flasharena · Table chart · 10 points Estimated from the imageHappyHorse by Alibaba ranks #2 in Text-to-Videoarena · Table chart · 7 points Estimated from the imageVideo Arenaarena · Bar chart · 13 points Estimated from the imagePareto Frontier: Text / Overall / Style Control Cost vs Scorearena · Scatter plot · 18 points Estimated from the imageWeighted average cost (USD) per Artificial Analysis Intelligence Index: Gemini 3.1 Pro Preview to Gemini 4 Argonartificial-analysis · Benchmark bar chart · 12 points Estimated from the imageWan2.6-t2v #1 Chinese lab in Video Arenaarena · Table chart · 30 points Estimated from the imageArena Gemini 3 Pro vs. 3.1 Pro Category Rankingsarena · Table chart · 28 points Estimated from the imageCost to Run Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 57 points Estimated from the imageMATH Level 5 pass@1 for Mistral Small 3 compared to other modelsepoch-ai · Bar chart · 5 points Estimated from the imageCost to Run Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 132 points Estimated from the imageWan 2.5 t2i preview ranks #5 in Text-to-Imagearena · Table chart · 60 points Estimated from the imageOutput Tokens per Secondartificial-analysis · Bar chart · 33 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Grouped bar chart · 20 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Grouped bar chart · 22 points Estimated from the imageOutput Tokens used Run all evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 90 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Bar chart · 41 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Grouped bar chart · 42 points Estimated from the imageGDPval-AA Leaderboardartificial-analysis · Benchmark bar chart · 49 points Estimated from the imagePareto Frontierarena · Scatter plot · 6 points Estimated from the imageElo scores from blind preference votes in our Image Arenaartificial-analysis · Bar chart · 20 points Estimated from the imageAgentic Performance (τ-Verb)artificial-analysis · Bar chart · 26 points Estimated from the imageAA-Briefcase Elo and GDPval-AI v2 Leaderboardartificial-analysis · Benchmark bar chart · 41 points Estimated from the imageArtificial Analysis Word Error Rate (AA-WER) Index by Open Weights / Proprietaryartificial-analysis · Benchmark bar chart · 40 points Estimated from the imageArtificial Analysis Image Arena Leaderboardartificial-analysis · Table chart · 22 points Estimated from the imageOutput Tokens per Intelligence Indexartificial-analysis · Benchmark bar chart · 50 points Estimated from the imageFrontend Gemini-3.6 Code Arena: Rank #12arena · Bar chart · 26 points Estimated from the imageText-to-Image Arenaarena · Bar chart · 15 points Estimated from the imageArena Preference Elo vs. Cost per Hour of Input Audioartificial-analysis · Scatter plot · 21 points Estimated from the imageText-to-Image Arena Nano Banana 2 Lite Pareto Frontierarena · Line chart · 7 points Estimated from the imageVeo-3.1-1080p ranks #2 & 5 in Video Arenaarena · Table chart · 20 points Estimated from the imageArtificial Analysis Speech to Speech Index vs. Cost per Hour of Input Audioartificial-analysis · Scatter plot · 12 points Estimated from the imageOutput Tokens per Secondartificial-analysis · Bar chart · 30 points Estimated from the imageSpeech to Speech Arena Eloartificial-analysis · Bar chart · 19 points Estimated from the imageHighlights from Q3, our first-ever $100B revenue quarterGoogle · Table chart · 10 points Estimated from the imageQ1 Earnings Highlights: AI investments and full stack approach are driving performance across our businessGoogle · Table chart · 8 points Estimated from the imageQ4 Earnings Highlights: Annual revenues exceeded $400B for the first timeGoogle · Table chart · 7 points Estimated from the imageGemini Model Capability ComparisonGoogle · Table chart · 39 points Estimated from the imageGemini 3 Deep ThinkGoogle · Benchmark bar chart · 15 points Estimated from the imageKnowledge workGoogle · Table chart · 12 points Estimated from the imageChatbot Arena Elo ScoreGoogle · Bar chart · 7 points Estimated from the imageTrending Food ItemsGoogle · Table chart · 6 points Estimated from the imageDomestic FlightsGoogle · Table chart · 2 points Estimated from the imageGemini 3.8 Flash vs. Other ModelsGoogle · Table chart · 102 points Estimated from the imageAl momentum redefining what's possible across the companyGoogle · Table chart · 6 points Estimated from the imageWeather ConditionsGoogle · Table chart · 3 points

Meta 96

Estimated from the imageReplica 8.3Llama3.1 405B Human EvaluationMeta · Stacked bar chart · 9 points Estimated from the imageReplica 8.1AI Model Performance TrendMeta · Line chart · 4 points Estimated from the imageTerminal-Bench 2.1, DeepSWE 1.1, Meta Internal Coding BenchMeta · Benchmark bar chart · 17 points Estimated from the imageGoogle and Meta are Benchmaxxerssemianalysis · Table chart · 12 points Estimated from the imageBenchmark ComparisonMeta · Table chart · 13 points Estimated from the imageMCP Atlas vs GDPval-AA V2Meta · Benchmark bar chart · 14 points Estimated from the imageBenchmarkMeta · Table chart · 54 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 32 points Estimated from the imageReplica 7.1Artificial Analysis Intelligence Index v4.1artificial-analysis · Bar chart · 37 points Estimated from the imageBenchmark ComparisonMeta · Table chart · 78 points Estimated from the imageArtificial Analysis Intelligence Index v4.1.1artificial-analysis · Bar chart · 54 points Estimated from the imageSharp threshold for exact fit in high dimensionsMeta · Table chart · 10 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Table chart · 44 points Estimated from the imageReplica 7.8LMArena ELO score vs. costMeta · Scatter plot · 13 points Estimated from the imageHeld Out Codebase PerplexityMeta · Line chart · 12 points Estimated from the imageReplica 8.4Improvement With Self-RefinementMeta · Stacked bar chart · 6 points Estimated from the imageOut-of-Distribution Performance vs In-Distribution PerformanceMeta · Scatter plot · 8 points Estimated from the imageImprovement With Test-Time ComputeMeta · Line chart · 10 points Estimated from the imageDiarizationMeta · Bar chart · 6 points Estimated from the imageReplica 7.6MemoryBenchMeta · Bar chart · 5 points Estimated from the imageFocusBenchMeta · Bar chart · 5 points Estimated from the imageSI-SDR improvement vs Number of speakersMeta · Scatter plot · 26 points Estimated from the imageArtificial Analysis Intelligence Index v4.3.2artificial-analysis · Grouped bar chart · 47 points Estimated from the imageIntelligence Index vs. Cost per Intelligence Index Taskartificial-analysis · Scatter plot · 28 points Estimated from the imageNet Win Rate of SAM Audio vs. SoTA SeparationMeta · Grouped bar chart · 92 points Estimated from the imagePersonal and professional agents benchmark comparisonMeta · Table chart · 65 points Estimated from the imageModel Performance ComparisonMeta · Benchmark bar chart · 32 points Estimated from the imageDFlash Speculative Decoding PerformanceMeta · Benchmark bar chart · 6 points Estimated from the imageSAM 3.1 and SAM 3 with Realistic Workload on a Single November 2025 H100Meta · Line chart · 16 points Estimated from the imageCode Muse Spark 1.3 (xHigh) Ranked ~#10arena · Bar chart · 31 points Estimated from the imageAIRA3 - NVIDIA Nemotron Model Reasoning ChallengeMeta · Table chart · 39 points Estimated from the imageText-to-Image Arenaarena · Bar chart · 15 points Estimated from the imageArtificial Analysis Intelligence Index v4.1.1artificial-analysis · Bar chart · 40 points Estimated from the imageCircular network diagram of shift and generalisation attributesMeta · Table chart · 17 points Estimated from the imageCyber Misuse Refusal BenchmarkMeta · Bar chart · 4 points Estimated from the imageMultimodal IntelligenceMeta · Line chart · 4 points Estimated from the imageMishandling Personal InformationMeta · Bar chart · 4 points Estimated from the imageText-to-Video Arena Ranks #3arena · Bar chart · 12 points Estimated from the imageWideSearchMeta · Line chart · 24 points Estimated from the imageAA-WER Streaming Index - Final TranscriptionMeta · Bar chart · 9 points Estimated from the imageReplica 7.9Model Benchmark ComparisonMeta · Table chart · 98 points Estimated from the imageArtificial Intelligence Index incorporates 7 evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, AIME, MATH-500artificial-analysis · Bar chart · 16 points Estimated from the imageVibe Code Bench v1.1Meta · Benchmark bar chart · 4 points Estimated from the imageExact vs Not exactMeta · Table chart · 10 points Estimated from the imageHumanity's Last Exam (With Tools)Meta · Line chart · 10 points Estimated from the imageMeta Internal Coding BenchMeta · Bar chart · 5 points Estimated from the imageV-JEPA 2 vs V-JEPA 2.1 vs Previous SOTAMeta · Benchmark bar chart · 24 points Estimated from the imageLlama 3.1Meta · Bar chart · 3 points Estimated from the imageModel Performance ComparisonMeta · Benchmark bar chart · 15 points Estimated from the imageMuse Image: Gap to Use Case Frontier (Text to Image)artificial-analysis · Table chart · 30 points Estimated from the imageMuse Image: Gap to Editing Action Frontier (Image Editing)artificial-analysis · Table chart · 21 points Estimated from the imageArtificial Analysis underlying Openness Index: Componentsartificial-analysis · Stacked bar chart · 80 points Estimated from the imageCost per Intelligence Index Taskartificial-analysis · Benchmark bar chart · 35 points Estimated from the imageArtificial Analysis Coding Agent Index vs. average total tokens per taskartificial-analysis · Scatter plot · 16 points Estimated from the imageArtificial Analysis Coding Agent Indexartificial-analysis · Benchmark bar chart · 31 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 23 points Estimated from the imageCritE Scoreartificial-analysis · Grouped bar chart · 25 points Estimated from the imageOutput Speed vs. Output Price: Llama 4 Maverick Providersartificial-analysis · Scatter plot · 11 points Estimated from the imageMulti-Meta Image Edit Arena Ranks #2arena · Benchmark bar chart · 15 points Estimated from the imageHarvey LAB-AA v1.1: All-Pass Rateartificial-analysis · Grouped bar chart · 21 points Estimated from the imageOutput Tokens per Second: Llama; Higher is Maverickartificial-analysis · Bar chart · 9 points Estimated from the imageOutput Speed vs. Price: Llama 4 Scoutartificial-analysis · Scatter plot · 5 points Estimated from the imageMuse Image: Gap to Capability Frontier (Text to Image)artificial-analysis · Table chart · 27 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Benchmark bar chart · 59 points Estimated from the imageOutput Speed vs. Price: Llama 4 Scout Providersartificial-analysis · Scatter plot · 10 points Estimated from the imageContext Window: Tokens Limit: Llama 4 Maverick Providersartificial-analysis · Bar chart · 10 points Estimated from the imageArtificial Analysis Coding Agent Index vs. average pay-per-token API cost per task (USD)artificial-analysis · Scatter plot · 26 points Estimated from the imageOutput Tokens Used to Run Artificial Analysis Intelligence Indexartificial-analysis · Stacked bar chart · 50 points Estimated from the imagePricing per 1M Tokens: Llama 4 Scout Providersartificial-analysis · Grouped bar chart · 20 points Estimated from the imageImage Meta Muse Image Ranks #2arena · Bar chart · 15 points Estimated from the imageCost per Taskartificial-analysis · Bar chart · 15 points Estimated from the imagePresentation Elo vs. Rubric Score (%)artificial-analysis · Scatter plot · 26 points Estimated from the imageOutput Tokens per Second: Llama 4 Maverick Providersartificial-analysis · Bar chart · 9 points Estimated from the imageGDPval-AAv2 Leaderboardartificial-analysis · Benchmark bar chart · 37 points Estimated from the imageTraining compute (FLOP) of notable open modelsepoch-ai · Scatter plot · 27 points Estimated from the imageOutput Speed vs. Price: Llama 4 Maverick Providersartificial-analysis · Scatter plot · 4 points Estimated from the imageAgent Arena Muse Spark 1.3 (Max)arena · Table chart · 6 points Estimated from the imageAgent Arena Spark 1.3 (Max): Ranked #13arena · Bar chart · 25 points Estimated from the imageτ³-Benchmarking: Scoreartificial-analysis · Bar chart · 39 points Estimated from the imageUSD per 1M Tokens: Input and Output Prices: Llama 4 Maverick Providersartificial-analysis · Grouped bar chart · 18 points Estimated from the imageAgent Arena Muse Spark 1.2 (xHigh)arena · Table chart · 6 points Estimated from the imageAgent Arena Ranked #25arena · Bar chart · 28 points Estimated from the imageText Arena - Open Weight Rank #24arena · Bar chart · 31 points Estimated from the imageMuse Image: Gap to Use Case Frontier (Image Editing)artificial-analysis · Table chart · 30 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Benchmark bar chart · 275 points Estimated from the imageCode Muse Arena - Open Weight Rank #26arena · Benchmark bar chart · 30 points Estimated from the imagePareto Frontier for Code Arenaarena · Scatter plot · 29 points Estimated from the imageGPQA Accuracies for Leading Modelsepoch-ai · Grouped bar chart · 8 points Estimated from the imageContext Window: Tokens Limit; Llama 4 Higher is better Providersartificial-analysis · Bar chart · 10 points Estimated from the imageMETA AI MODEL RESULTSMeta · Table chart · 5 points Estimated from the imageOpen Materials 2024 (OMat24)Meta · Line chart · 18 points Estimated from the imageBuilt with LlamaMeta · Table chart · 10 points Estimated from the imageAIRA3 - NVIDIA Nemotron Model Reasoning ChallengeMeta · Line chart · 97 points Estimated from the imageBenchmarkMeta · Table chart · 47 points Estimated from the imageLive Evaluation ResultsMeta · Benchmark bar chart · 26 points Estimated from the imageWild Artifact BenchMeta · Scatter plot · 21 points

xAI 108

Estimated from the imageReplica 8.1Humanity's Last ExamxAI · Bar chart · 7 points Estimated from the imageCaseLaw BenchmarkxAI · Table chart · 6 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Bar chart · 17 points Estimated from the imageCursorBench 4.0 scorexAI · Line chart · 17 points Estimated from the imageGrok-4.20-Beta1 ranks #1 in Search Arenaarena · Table chart · 40 points Estimated from the imageReplica 5.3Artificial Analysis Intelligence Index v4.1.1artificial-analysis · Bar chart · 29 points Estimated from the imageReplica 8.1Artificial Analysis Intelligence Index v4.1artificial-analysis · Grouped bar chart · 30 points Estimated from the imageAI Benchmark ComparisonxAI · Benchmark bar chart · 46 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Bar chart · 40 points Estimated from the imageAI Model Benchmarksartificial-analysis · Benchmark bar chart · 120 points Estimated from the imagexAI Grok-3 #1 in Arenaarena · Table chart · 13 points Estimated from the imageReplica 8.1Intelligence vs. PricexAI · Scatter plot · 17 points Estimated from the imageScaling HLE - Test timexAI · Scatter plot · 28 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Bar chart · 35 points Estimated from the imageIntelligence vs. PricexAI · Scatter plot · 9 points Estimated from the imageCode Arena: Ranked #3arena · Bar chart · 27 points Estimated from the imager²-bench TelecomxAI · Scatter plot · 2 points Estimated from the imageAA-WER Streaming Index - Final Transcription after detected End of Speechartificial-analysis · Grouped bar chart · 41 points Estimated from the imageHallucination Rate vs FActScorexAI · Benchmark bar chart · 4 points Estimated from the imageGemini 2.5 Flash Leaderboardarena · Table chart · 45 points Estimated from the imageArtificial Analysis Coding Agent Index vs. Cost per Taskartificial-analysis · Benchmark bar chart · 22 points Estimated from the imageGrok 4.1 vs. previous GrokxAI · Table chart · 1 points Estimated from the imageBig Bench Audio: Intelligence vs LatencyxAI · Scatter plot · 1 points Estimated from the imageModel PerformancexAI · Scatter plot · 1 points Estimated from the imageArtificial Analysis Speech to Speech Indexartificial-analysis · Benchmark bar chart · 35 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 19 points Estimated from the imageCost per minutexAI · Bar chart · 5 points Estimated from the imageGrok-3 is also #1 across all categoriesarena · Table chart · 120 points Estimated from the imageArtificial Analysis Intelligence Index vs. Priceartificial-analysis · Scatter plot · 15 points Estimated from the imageBerkeley Function Calling v4 BenchmarkxAI · Scatter plot · 6 points Estimated from the imageSpeech Think-to-Speech IndexxAI · Bar chart · 8 points Estimated from the imageArtificial Analysis Cyber Indexartificial-analysis · Benchmark bar chart · 48 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Bar chart · 35 points Estimated from the imageEQ-Emotional Intelligence Benchmark - Elo (Normalized)xAI · Bar chart · 8 points Estimated from the imageAgent Arenaarena · Bar chart · 19 points Estimated from the imageSpeech Transcription: Accuracy vs. Price (Streaming)xAI · Scatter plot · 11 points Estimated from the imageInstruction followingxAI · Bar chart · 8 points Estimated from the imagexAI Grok-4 tied for #3 in Text Arenaarena · Table chart · 33 points Estimated from the imageText-to-Image Arena Top 20arena · Bar chart · 20 points Estimated from the imageGrok-4-fast-search lands at #1arena · Table chart · 43 points Estimated from the imageReplica 8.1AI Voice Model ComparisonxAI · Table chart · 19 points Estimated from the imageLMArena Text LeaderboardxAI · Bar chart · 16 points Estimated from the imageClaude Opus-4-6 & Sonnet 4.6 ranks #1 & #7 in Search Arenaarena · Table chart · 19 points Estimated from the imageCreative Writing v3xAI · Bar chart · 7 points Estimated from the imageArtificial Analysis Intelligence Index (Version 2, released Feb '25); Higher is betterartificial-analysis · Bar chart · 21 points Estimated from the imageModel Performance ComparisonxAI · Benchmark bar chart · 6 points Estimated from the imageTelephony (8 kHz)xAI · Grouped bar chart · 32 points Estimated from the imageGemini 2.5 Pro #1 on Arena with the largest score jump ever (+40)!arena · Table chart · 17 points Estimated from the imageArtificial Analysis Text to Video Leaderboardartificial-analysis · Table chart · 30 points Estimated from the imageOutput Tokens used Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 63 points Estimated from the imageModel Rankingsarena · Table chart · 7 points Estimated from the imageIntelligence Analysis vs. Cost to Run Artificial Analysis Intelligence Indexartificial-analysis · Scatter plot · 20 points Estimated from the imageAverage time to First audio (seconds) on Big Bench Audio dataset - Lower is betterartificial-analysis · Bar chart · 29 points Estimated from the imageAgent Arenaarena · Bar chart · 6 points Estimated from the imageArtificial Analysis Word Error Rate Index (Non-streaming) vs Priceartificial-analysis · Scatter plot · 32 points Estimated from the imagePrice per 1000 minutes of audioartificial-analysis · Bar chart · 38 points Estimated from the imageCode Arena: WebDev Ranked #10arena · Bar chart · 24 points Estimated from the imageIntelligence Analysis Index vs. Cost per Task (USD)artificial-analysis · Scatter plot · 28 points Estimated from the imageTime per Intelligence Index Taskartificial-analysis · Bar chart · 26 points Estimated from the imageArtificial Reasoning, Conversational Speech to Speech and Agentic Index Performance by Datasetartificial-analysis · Benchmark bar chart · 42 points Estimated from the imageOutput Tokens per Intelligence Indexartificial-analysis · Benchmark bar chart · 81 points Estimated from the imageOutput Tokens Speed per Second; Higher is betterartificial-analysis · Bar chart · 21 points Estimated from the imageAA-WER Streaming Index vs. Time to Final Transcriptionartificial-analysis · Scatter plot · 24 points Estimated from the imageAA-WER: % of words transcribed incorrectly on the Earnings22-Cleaned-AA-v1 Datasetartificial-analysis · Bar chart · 36 points Estimated from the imageAA-Briefcase Elo vs. Cost per Taskartificial-analysis · Scatter plot · 42 points Estimated from the imageFrontierMath accuracy for Grok-3 and Grok-3 mini compared to leading modelsepoch-ai · Bar chart · 9 points Estimated from the imageArtificial Analysis Word Error Rate (AA-WER) Indexartificial-analysis · Bar chart · 33 points Estimated from the imageGrok Imagine Image 2.0: Gap to Capability Frontier (Text to Image)artificial-analysis · Table chart · 30 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 39 points Estimated from the imageTotal Training FLOPs by Modelepoch-ai · Line chart · 6 points Estimated from the imageOutput Tokens Used to Run Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 46 points Estimated from the imageGrok Imagine Image 2.0: Gap to Editing Action Frontier (Image Editing)artificial-analysis · Table chart · 21 points Estimated from the imageGrok Imagine Video 1.5 Lite: Gap to Capability Frontier (AA-Video-T2V v2.0)artificial-analysis · Table chart · 30 points Estimated from the imageAA-Briefcase Elo vs. Time per Taskartificial-analysis · Scatter plot · 47 points Estimated from the imageGDPval-AA v2 Leaderboardartificial-analysis · Bar chart · 32 points Estimated from the imageTraining compute of notable modelsepoch-ai · Scatter plot · 7 points Estimated from the imageAgent Arena: Grok over timearena · Line chart · 8 points Estimated from the imageEnd-to-End Output 500 Response Timeartificial-analysis · Grouped bar chart · 58 points Estimated from the imageImage-to-Video Arenaarena · Bar chart · 16 points Estimated from the imageGrok Imagine Video 1.5 Lite: Quality vs Speed Frontier (AA-Video-T2V-Silent v2.0)artificial-analysis · Scatter plot · 9 points Estimated from the imageImage Edit Arena Top 20arena · Bar chart · 19 points Estimated from the imageAA-WER Streaming Index (First Partial) vs. Time to First Partial Transcription After Speech Endartificial-analysis · Scatter plot · 21 points Estimated from the imageReplica 5.3Artificial Analysis Intelligence Index V4.3artificial-analysis · Bar chart · 29 points Estimated from the imageArtificial Analysis Coding Agent Indexartificial-analysis · Benchmark bar chart · 69 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 36 points Estimated from the imageGrok Imagine Video 1.5 Lite: Gap to Use Case Frontierartificial-analysis · Table chart · 30 points Estimated from the imageGrok Imagine Image 2.0: Gap to Use Case Frontier (Text to Image)artificial-analysis · Table chart · 20 points Estimated from the imageText-to-Video Arenaarena · Bar chart · 15 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Grouped bar chart · 27 points Estimated from the imageArtificial Analysis Coding Agent Index (Composite average pass@1 across DeepSWE, Terminal-Bench v2, and SWE-Atlas-QnA)artificial-analysis · Grouped bar chart · 20 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Benchmark bar chart · 32 points Estimated from the imageCode Arena: WebDevarena · Bar chart · 15 points Estimated from the imageAA-WER Streaming Index (First Partial) vs. Time to First Partial Transcription After Speech Endartificial-analysis · Scatter plot · 32 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Grouped bar chart · 42 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Bar chart · 100 points Estimated from the imageDeepSWE v1.1 Benchmark Scoreartificial-analysis · Grouped bar chart · 19 points Estimated from the imageGrok Imagine Image 2.0: Gap to Use Case Frontier (Image Editing)artificial-analysis · Table chart · 20 points Estimated from the imageLanguage Model Inference Priceartificial-analysis · Line chart · 80 points Estimated from the imageCost per complete fixed 40-question Big Audio Bench (Big Bench Audio subset based on length of input audio, normalized to hourly basis - Lower is better)artificial-analysis · Bar chart · 21 points Estimated from the imageSWE-Atlas-QnA Benchmark ScorexAI · Bar chart · 14 points Estimated from the imageToken usage by modelxAI · Table chart · 110 points Estimated from the imageToken usage LeaderboardxAI · Stacked bar chart · 66 points Estimated from the imageTop chartsxAI · Table chart · 6 points Estimated from the imageTSLA Tesla, Inc.xAI · Line chart · 5 points Estimated from the imageCategoriesxAI · Stacked bar chart · 9 points Estimated from the imageVEHICLE SUMMARYxAI · Table chart · 44 points Estimated from the imageFrontierCode 1.1 Main Score vs CostxAI · Scatter plot · 16 points Estimated from the imageVoting intentionxAI · Bar chart · 10 points

DeepSeek 97

Partly publisher's numbersReplica 8.1Accuracy / Percentile (%)DeepSeek · Benchmark bar chart · 30 points Estimated from the imageGlobal KV Cache Per Token (Bytes)DeepSeek · Bar chart · 4 points Estimated from the imageReplica 6.3AI Model Benchmark Performance Comparisonqwen · Benchmark bar chart · 25 points Partly publisher's numbersReplica 8.3Accuracy / Percentile (%)DeepSeek · Benchmark bar chart · 30 points Partly publisher's numbersAccuracy / Percentile (%)DeepSeek · Benchmark bar chart · 36 points Estimated from the imageReplica 9.0AI Model Benchmarks ComparisonDeepSeek · Benchmark bar chart · 20 points Estimated from the imageReplica 7.4AI Model Benchmarks Comparisonqwen · Benchmark bar chart · 25 points Estimated from the imageReplica 9.0Model Performance Comparison Across Benchmarksgithub:zai-org/GLM-4 · Benchmark bar chart · 30 points Partly publisher's numbersReplica 8.2Accuracy Percentile by Model and BenchmarkDeepSeek · Benchmark bar chart · 25 points Estimated from the imageReplica 8.6Artificial Analysis Intelligence Index incorporates 7 evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, AIME, MATH-500artificial-analysis · Bar chart · 16 points Estimated from the imageArtificial Analysis Intelligence Index (Non-Reasoning Models)artificial-analysis · Bar chart · 17 points Partly publisher's numbersAccuracy / Pass@1 (%)DeepSeek · Benchmark bar chart · 28 points Estimated from the imageReplica 7.3Model Performance Comparison Across Benchmarksgithub:zai-org/GLM-4 · Benchmark bar chart · 50 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 63 points Estimated from the imageBenchmark Comparison of Reasoning and Agentic CapabilitiesDeepSeek · Benchmark bar chart · 44 points Estimated from the imageCost Per Million Tokens vs Token PositionDeepSeek · Line chart · 20 points Estimated from the imagePressure Testing DeepSeek-V3 128K Context via "Needle In A HayStack"DeepSeek · Table chart · 15 points Estimated from the imageDeepSeek Model Performance ComparisonDeepSeek · Line chart · 30 points Estimated from the imagePressure Testing DeepSeek-Coder-V2 Base 128K Context via "Needle In A HayStack"DeepSeek · Table chart · 11 points Estimated from the imagedeepseek-r1-lite-preview AIME accuracy with test-time compute scalingDeepSeek · Line chart · 17 points Estimated from the imagePass Rate ComparisonDeepSeek · Benchmark bar chart · 14 points Estimated from the imageAgent DeepSeek-V4.1-Flash (Max): Ranked #3 Openarena · Grouped bar chart · 15 points Estimated from the imageAverage PerformanceDeepSeek · Table chart · 9 points Estimated from the imageInference costs of DeepSeek-V3.1-Terminus and DeepSeek-V3.2-ExpDeepSeek · Line chart · 20 points Estimated from the image(a) Compression on Fox benchmarkDeepSeek · Table chart · 28 points Estimated from the imageCode DeepSeek Arena Top3 Openarena · Bar chart · 15 points Estimated from the imagePass Rate ComparisonDeepSeek · Benchmark bar chart · 30 points Estimated from the imageModel Performance on Three BenchmarksDeepSeek · Benchmark bar chart · 15 points Partly publisher's numbersAIME 2024DeepSeek · Bar chart · 6 points Estimated from the imageAgent DeepSeekArena-V4-Flash (High)arena · Bar chart · 30 points Estimated from the imageProgramming Language Performance ComparisonDeepSeek · Table chart · 48 points Estimated from the imageBenchmark PerformancesDeepSeek · Table chart · 45 points Estimated from the imageLeaderboardDeepSeek · Scatter plot · 23 points Estimated from the imageMMLU Redux ZeroEval Score VS Input API Price ($ / 1M Tokens)DeepSeek · Scatter plot · 13 points Estimated from the imageArtificial Analysis Intelligence Index v2artificial-analysis · Bar chart · 22 points Estimated from the imageDeepSeek Model BenchmarksDeepSeek · Benchmark bar chart · 12 points Estimated from the imageHuman evaluationsDeepSeek · Benchmark bar chart · 20 points Partly publisher's numbersAccuracy / Pass@1 (%)DeepSeek · Benchmark bar chart · 31 points Estimated from the imageBenchmark PerformanceDeepSeek · Table chart · 36 points Estimated from the imageAI Model Benchmark Performance ComparisonDeepSeek · Benchmark bar chart · 20 points Estimated from the imageFrontend DeepSeek-V4 Flash Arena (High)arena · Bar chart · 19 points Estimated from the imageDeepSeek V4 Flash 0731 debuts as the second-strongest open-weights modelepoch-ai · Benchmark bar chart · 11 points Estimated from the imagePareto Frontier for Frontend Code Arenaarena · Line chart · 6 points Estimated from the imageInterpolated Token Throughput per GPU (tok/s/gpu)semianalysis · Line chart · 85 points Estimated from the imageIntelligence Parameters vs. Active Parametersartificial-analysis · Scatter plot · 15 points Estimated from the imageAI Model Accuracy Comparison Across BenchmarksDeepSeek · Benchmark bar chart · 42 points Estimated from the imageMATH Top@1 Accuracy over TimeDeepSeek · Line chart · 7 points Partly publisher's numbersReplica 7.9Accuracy / Percentile (%) by BenchmarkDeepSeek · Benchmark bar chart · 30 points Estimated from the imageAverage Performance vs Activated ParametersDeepSeek · Scatter plot · 10 points Partly publisher's numbersAIME 2024DeepSeek · Bar chart · 8 points Estimated from the imageReplica 7.8Input/Output Pricing for o1-Class Inference Models ($/1M Tokens)DeepSeek · Grouped bar chart · 12 points Estimated from the imageArtificial Analysis Intelligence Index v4.3artificial-analysis · Table chart · 24 points Estimated from the imageAgent DeepSeek-V4-Pro (High)arena · Bar chart · 16 points Estimated from the imageArtificial Analysis Intelligence Index v4.1 by Open Weights / Proprietaryartificial-analysis · Benchmark bar chart · 56 points Estimated from the imageWin Rates Across Website Designarena · Grouped bar chart · 32 points Estimated from the imageGDPval-AA Models Leaderboardartificial-analysis · Bar chart · 16 points Estimated from the imageAA-Briefcase rubric Score from binary rubric checks (%)artificial-analysis · Bar chart · 30 points Estimated from the imageGDPval-AA Leaderboardartificial-analysis · Benchmark bar chart · 52 points Estimated from the imageGDPval-AA v2artificial-analysis · Grouped bar chart · 35 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 28 points Estimated from the imageText DeepSeek:V4-Top15 (Max)arena · Bar chart · 16 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 56 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Grouped bar chart · 25 points Estimated from the imageJev Routerarena · Bar chart · 11 points Estimated from the imageCode DeepSeek Arena: V4-Pro (Max) Top15arena · Bar chart · 15 points Estimated from the imageOutput Tokens per Second: DeepSeek V3.2 Exp; Higher is better; 1,000 Input Providersartificial-analysis · Bar chart · 4 points Estimated from the imageArtificial Analysis Coding Indexartificial-analysis · Bar chart · 32 points Estimated from the imageElo rating-AA for performance on real-world work tasksartificial-analysis · Bar chart · 28 points Estimated from the imageTop 7 Open Models: February 2026 in Code Arenaarena · Bar chart · 7 points Partly publisher's numbersIntelligence Analysis vs. Intelligence Index; Output Tokens Used in Artificial Analysis Intelligence Indexartificial-analysis · Scatter plot · 8 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 27 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Grouped bar chart · 24 points Estimated from the imageGDPval-AA v2 Leaderboardartificial-analysis · Bar chart · 35 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Bar chart · 100 points Estimated from the imageAutomationBench-AA: Score vs. Cost per Taskartificial-analysis · Scatter plot · 22 points Estimated from the imageOutput Tokensartificial-analysis · Benchmark bar chart · 58 points Estimated from the imageText Arena Score by Categoryarena · Table chart · 70 points Estimated from the imageIntelligence vs. Cost to Run Artificial Intelligence Indexartificial-analysis · Scatter plot · 32 points Estimated from the imageAgent DeepSeek-V4-Pro (High)arena · Benchmark bar chart · 12 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Table chart · 34 points Estimated from the imagePareto Frontier for Agent Arenaarena · Scatter plot · 9 points Estimated from the imageOutput Reasoning Tokens Used to Run Artificial Analysis Intelligence Indexartificial-analysis · Bar chart · 23 points Estimated from the imageOutput Tokens per Second: DeepSeek V3.1artificial-analysis · Bar chart · 5 points Estimated from the imageAgent DeepSeek Arena Pareto Frontierarena · Scatter plot · 9 points Estimated from the imageAgent DeepSeek-Arena-V4-Flash: Rank #21arena · Bar chart · 6 points Estimated from the imageWin Rates Across Game Developmentarena · Benchmark bar chart · 28 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 50 points Estimated from the imageCost per Intelligence Indexartificial-analysis · Benchmark bar chart · 130 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Benchmark bar chart · 59 points Estimated from the imageAA-Briefcase Eloartificial-analysis · Bar chart · 39 points Estimated from the imageDeepSeek-V4.1-Flash (Max) Pareto Frontier for Agent Arenaarena · Line chart · 14 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Benchmark bar chart · 107 points Estimated from the imageIntelligence vs. Total Parametersartificial-analysis · Scatter plot · 19 points Estimated from the imageDeepSeek V4 Pro (Max) Pareto Frontier for Code Arenaarena · Scatter plot · 8 points Estimated from the imageArtificial Analysis Intelligence Index by Open Weights / Proprietaryartificial-analysis · Bar chart · 24 points Estimated from the imageDeepSeek-V4 API New PricingDeepSeek · Table chart · 12 points Partly publisher's numbersBenchmark (Metric) ComparisonDeepSeek · Table chart · 111 points

Qwen 117

Estimated from the imageAI Model Benchmark Performance ComparisonQwen · Benchmark bar chart · 25 points Estimated from the imageReplica 7.8AI Model Benchmarks ComparisonQwen · Benchmark bar chart · 25 points Estimated from the imageModel Comparison Across BenchmarksQwen · Benchmark bar chart · 92 points Estimated from the imageQwen3.8-Max: Ranks #2arena · Grouped bar chart · 19 points Estimated from the imageModel BenchmarksQwen · Benchmark bar chart · 55 points Estimated from the imageModel Benchmark ComparisonQwen · Table chart · 58 points Estimated from the imageModel Benchmark ComparisonQwen · Table chart · 63 points Estimated from the imageModel Benchmarks ComparisonQwen · Benchmark bar chart · 85 points Estimated from the imageReplica 7.2Model Comparison BenchmarksQwen · Benchmark bar chart · 84 points Estimated from the imagePareto FrontierQwen · Scatter plot · 7 points Estimated from the imageModel Comparison Across BenchmarksQwen · Benchmark bar chart · 120 points Estimated from the imageCode Arena 3.7 Max (20260517) : #3arena · Bar chart · 19 points Estimated from the imageReplica 7.4Frontend3 Code Max: Ranks #4arena · Bar chart · 19 points Estimated from the imageReplica 7.7Code Arena: WebDevarena · Bar chart · 14 points Estimated from the imageQwen3.8-Max Pareto Frontier for Frontend Code Arenaarena · Line chart · 7 points Estimated from the imageRelative Prefill Throughput at 90% Cache Hit RateQwen · Line chart · 21 points Estimated from the imageASR Error Rates - Public TestQwen · Grouped bar chart · 29 points Estimated from the imageReplica 8.4Qwen3-Coder-Plus Benchmark ComparisonQwen · Benchmark bar chart · 8 points Estimated from the imageArtificial Analysis Intelligence IndexQwen · Bar chart · 43 points Estimated from the imageAverage Ranking vs. Environment ScalingQwen · Line chart · 14 points Estimated from the imageQwen3-Coder-Flash BenchmarksQwen · Table chart · 135 points Estimated from the imageQwen3-TTS-12Hz-1.7B ComparisonQwen · Table chart · 58 points Estimated from the imageReplica 7.6Model Performance ComparisonQwen · Benchmark bar chart · 25 points Estimated from the imageArtificial Analysis Intelligence Agentic IndexQwen · Bar chart · 31 points Estimated from the imageSWE-bench Performance Verified (%)Qwen · Scatter plot · 11 points Estimated from the imageVramQwen · Bar chart · 2 points Estimated from the imageModel Benchmarks ComparisonQwen · Benchmark bar chart · 79 points Estimated from the imageModel Performance Comparison Across BenchmarksQwen · Benchmark bar chart · 42 points Estimated from the imageQwen3-Coder BenchmarksQwen · Table chart · 135 points Estimated from the imageModel Performance Comparison Across BenchmarksQwen · Benchmark bar chart · 78 points Estimated from the imageQwen 3.6 Max (preview) Benchmark ComparisonQwen · Benchmark bar chart · 70 points Estimated from the imageTotal images processed on OpenRouterQwen · Stacked bar chart · 70 points Estimated from the imageQwen2.5-Omni Performance ComparisonQwen · Benchmark bar chart · 10 points Estimated from the imageModel Benchmarks ComparisonQwen · Benchmark bar chart · 66 points Estimated from the imageAI Model Benchmark ComparisonQwen · Benchmark bar chart · 25 points Estimated from the imageQwen3-Omni-Flash2025-12-01 vs Qwen3-Omni-Flash vs Qwen3-235B-A22B vs GPT-4o vs Gemini-2.5-FlashQwen · Table chart · 158 points Estimated from the imageAI Model Benchmark Performance ComparisonQwen · Benchmark bar chart · 25 points Estimated from the imageContext LengthQwen · Benchmark bar chart · 6 points Estimated from the imageQwen3.5-Omni-Plus Benchmark ComparisonQwen · Table chart · 76 points Estimated from the imageContent Consistency (↓) - TTS Multilingual Test SetQwen · Benchmark bar chart · 40 points Estimated from the imageModel Benchmarks ComparisonQwen · Table chart · 223 points Estimated from the imageContent Consistency (↓) - TTS Multilingual Test SetQwen · Grouped bar chart · 65 points Estimated from the imageModel Benchmarks ComparisonQwen · Benchmark bar chart · 69 points Estimated from the imagePerformance on Coding Agent BenchmarksQwen · Benchmark bar chart · 20 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 37 points Estimated from the imageQwen-Image-2.1: Gap to Open Weights Use Case Frontierartificial-analysis · Table chart · 20 points Estimated from the imageQwen-RobotNav Benchmark ResultsQwen · Benchmark bar chart · 57 points Estimated from the imageChatbot Arena Category: Overall vs Creative Writingarena · Line chart · 12 points Estimated from the imageTotal Tokens per $1 TCO (Owning - Hyperscaler) vs. P90 Interactivitysemianalysis · Scatter plot · 66 points Estimated from the imageQwen-Image-2.1: Gap to Editing Action Frontier (AA-Image-Editing v2.0)artificial-analysis · Table chart · 21 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Table chart · 30 points Estimated from the imageOutput Tokens per Second: Qwen2.5 72Bartificial-analysis · Bar chart · 7 points Estimated from the imageWan 2.5-i2i-Preview ranks #21 in Image Editarena · Table chart · 18 points Estimated from the imageQwen3-ASR-1.7B vs Other ASR ModelsQwen · Table chart · 90 points Estimated from the imageIntelligence Analysis vs. Cost to Run Intelligence Indexartificial-analysis · Scatter plot · 31 points Estimated from the imageQwen2.5-Omni vs Other Models Across BenchmarksQwen · Benchmark bar chart · 21 points Estimated from the imageText-to-Image Model Elo LeaderboardQwen · Table chart · 55 points Estimated from the imageQwen-Image-2.1: Gap to Use Case Frontier (AA-Image-T2I v2.0)artificial-analysis · Table chart · 30 points Estimated from the imageTime to First Audioartificial-analysis · Bar chart · 13 points Estimated from the imageIntelligence vs. Active Parameters: Qwen3 Modelsartificial-analysis · Scatter plot · 16 points Estimated from the imageModel Performance ComparisonQwen · Benchmark bar chart · 48 points Estimated from the imageWan #3 i2v preview ranks #3 in Image-to-Videoarena · Table chart · 48 points Estimated from the imagePareto Frontier for Code Arena: WebDevarena · Line chart · 11 points Estimated from the imageOutput Tokens per Second, Code Oriented Prompt: Qwen2.5 Coder 32Bartificial-analysis · Bar chart · 5 points Estimated from the imageQwen-Image-2.1: Gap to Open Weights Editing Action Frontierartificial-analysis · Table chart · 21 points Estimated from the imageQwen-Image-2.1: Gap to Open Weights Capability Frontier (AA-Image-T2I v2.0)artificial-analysis · Table chart · 27 points Estimated from the imageQwen3-VL Benchmark ComparisonQwen · Table chart · 70 points Estimated from the imageContent Consistency (↓) - TTS Multilingual Test SetQwen · Grouped bar chart · 44 points Estimated from the imageText-to-Image-2.1 Arena: Ranked #1 Openarena · Bar chart · 10 points Estimated from the imageCompletion Time by Model and Systemartificial-analysis · Table chart · 14 points Estimated from the imageImage-to-WebDev Arena: Open Qwen-3.8-27B: Ranked #1arena · Bar chart · 10 points Estimated from the imageModel ComparisonQwen · Table chart · 209 points Estimated from the imageOutput Tokens per Second: Qwen3 235B 2507 (Reasoning) Providersartificial-analysis · Bar chart · 6 points Estimated from the image(a) Task & Scene GeneralizationQwen · Grouped bar chart · 8 points Estimated from the imageArtificial Analysis Intelligence Index v4.0, <40B Total Parametersartificial-analysis · Bar chart · 23 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Bar chart · 25 points Estimated from the imageQwen-Image-Edit model. Ranks the #6 open overall.arena · Table chart · 9 points Estimated from the imagePareto Frontier for Code Arena: WebDevarena · Line chart · 10 points Estimated from the imageQwen-Image-2.1: Gap to Open Weights Use Case Frontier (AA-Image-Editing v2.0)artificial-analysis · Table chart · 30 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Bar chart · 42 points Estimated from the imageAverage Score (16K max context) vs. Output Tokensartificial-analysis · Scatter plot · 21 points Estimated from the imageQwen3-Max-Instruct scores much higher on AA-SimpleQA Omniscience Verified would predict its score onepoch-ai · Scatter plot · 13 points Estimated from the imageText-to-Image Arenaarena · Bar chart · 14 points Estimated from the imageCross-Harness GeneralizationQwen · Benchmark bar chart · 12 points Estimated from the imageInstruction Following (↑) - InstructTTS EvalQwen · Benchmark bar chart · 24 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 28 points Estimated from the imageImage Edit Model Rankingsarena · Table chart · 15 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Bar chart · 43 points Estimated from the imageIntelligence Analysis vs. Total Intelligence Indexartificial-analysis · Scatter plot · 26 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 41 points Estimated from the imageIntelligence vs. Active Parameters; Artificial Analysis Intelligence Indexartificial-analysis · Scatter plot · 29 points Estimated from the imageQwen-Image-2.1: Gap to Capability Frontier (AA-Image-T2I v2.0)artificial-analysis · Table chart · 18 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Grouped bar chart · 22 points Estimated from the imageQwen3-ASR-0.6B vs Other ASR ModelsQwen · Table chart · 19 points Estimated from the imageModel ComparisonQwen · Table chart · 135 points Estimated from the imageTotal Tokens per $1 TCO (Owning - Hyperscaler Inference) vs. P90 Time To First Tokensemianalysis · Scatter plot · 47 points Estimated from the imageOutput Tokens per Second: Qwen3 Coder 480B Providersartificial-analysis · Bar chart · 10 points Estimated from the imageQwen3-VL vs Other ModelsQwen · Table chart · 148 points Estimated from the imageQwen Image 2.0 Pro ranks #9 for Image Arenaarena · Table chart · 10 points Estimated from the imageText-to-Image 3.0-Pro vs. 2.0-Pro Arena Score by Categoryarena · Table chart · 2 points Estimated from the imageIntelligence vs. Output Tokens Used in Artificial Analysis Intelligence Indexartificial-analysis · Scatter plot · 26 points Estimated from the imageImage Edit Arenaarena · Bar chart · 10 points Estimated from the imageQwen3-VL 235B-A22B Instruct vs Other ModelsQwen · Bar chart · 42 points Estimated from the imageQwen-Image-2.1: Gap to Use Case Frontier (AA-Image-Editing v2.0)artificial-analysis · Table chart · 20 points Estimated from the imageQwen Image 2512 is #2 open and #13 overall in Text-to-Imagearena · Table chart · 36 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 27 points Estimated from the imageASR Domain Term Recognition AccuracyQwen · Table chart · 27 points Estimated from the imageGPU utilization (%)Qwen · Table chart · 50 points Estimated from the imageMobileEval BenchmarksQwen · Table chart · 5 points Estimated from the imageModel Performance ComparisonQwen · Benchmark bar chart · 22 points Estimated from the image(1) Building the Foundation for Agentic Environment SimulationQwen · Grouped bar chart · 19 points Estimated from the imageQwen3-Omni ArchitectureQwen · Table chart · 16 points Estimated from the imageBest-of-8 Accuracy vs ProcessBench Mean F1Qwen · Benchmark bar chart · 24 points Estimated from the imageQwen2.5-Coder-32BQwen · Table chart · 55 points Estimated from the imageAI model leaderboardQwen · Table chart · 70 points Estimated from the imageAgent: Faster, More Cost-EffectiveQwen · Bar chart · 2 points Estimated from the imageText ArenaQwen · Table chart · 58 points

Kimi (Moonshot AI) 85

Estimated from the imageReplica 9.0Model Performance ComparisonKimi (Moonshot AI) · Benchmark bar chart · 15 points Estimated from the imageModel BenchmarksKimi (Moonshot AI) · Benchmark bar chart · 47 points Estimated from the imageReplica 7.9Coding and Agents BenchmarksKimi (Moonshot AI) · Benchmark bar chart · 24 points Estimated from the imageModel Performance ComparisonKimi (Moonshot AI) · Benchmark bar chart · 38 points Estimated from the imageInternal Knowledge Work BenchKimi (Moonshot AI) · Benchmark bar chart · 9 points Estimated from the imageReplica 6.3Frontend Code Kimi-K3: Ranked #1arena · Bar chart · 23 points Estimated from the imageOmniDocBench Scores by CategoryKimi (Moonshot AI) · Benchmark bar chart · 44 points Estimated from the imageReplica 3.5General AgentsKimi (Moonshot AI) · Benchmark bar chart · 45 points Estimated from the imageCodingKimi (Moonshot AI) · Benchmark bar chart · 36 points Estimated from the imageFrontend Code Arena United States vs Chinaarena · Line chart · 23 points Estimated from the imageKimi Code-K2.7 Code vs Kimi-K2.6 : Performance vs TokensKimi (Moonshot AI) · Scatter plot · 6 points Estimated from the imageAgent Arena Top 25 Models - Kimi K3: Rank #4arena · Benchmark bar chart · 30 points Estimated from the imageReplica 7.1Artificial Analysis Intelligence Index v4.1artificial-analysis · Bar chart · 44 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Benchmark bar chart · 72 points Estimated from the imageKimi K2 Thinking vs GPT-5 vs Claude Sonnet 4.5 (Thinking) Benchmark ComparisonKimi (Moonshot AI) · Benchmark bar chart · 18 points Estimated from the imageMathVision Pass@1 vs Activated ParametersKimi (Moonshot AI) · Scatter plot · 11 points Estimated from the imageTop ModelsKimi (Moonshot AI) · Stacked bar chart · 10 points Estimated from the imaget-SNE of MCP tools by CategoryKimi (Moonshot AI) · Scatter plot · 34 points Estimated from the imageTime series dataKimi (Moonshot AI) · Line chart · 20 points Estimated from the imageProportion vs Stated ConfidenceKimi (Moonshot AI) · Line chart · 125 points Estimated from the imageLLM Inference on M3 Max - Performance EvolutionKimi (Moonshot AI) · Scatter plot · 9 points Estimated from the imagePer-step training loss curve of Kimi K2Kimi (Moonshot AI) · Line chart · 18 points Estimated from the imageFigure 1: Kimi K2 main resultsKimi (Moonshot AI) · Benchmark bar chart · 56 points Estimated from the imageKimi Code BenchKimi (Moonshot AI) · Bar chart · 3 points Estimated from the imageKimi K2.5 API PriceKimi (Moonshot AI) · Table chart · 9 points Estimated from the imageLLM Inference on M3 Max - Performance EvolutionKimi (Moonshot AI) · Scatter plot · 45 points Estimated from the imageτ²-Bench Telecom (Agentic Tool Use)Kimi (Moonshot AI) · Bar chart · 25 points Estimated from the imageKimi K2 Thinking vs GPT-5 vs Claude Sonnet 4.5 (Thinking)Kimi (Moonshot AI) · Benchmark bar chart · 18 points Estimated from the imageOutput SpeedKimi (Moonshot AI) · Bar chart · 2 points Estimated from the imageKimi Code BenchKimi (Moonshot AI) · Bar chart · 2 points Estimated from the imageKimi K2.6 Agent via kimi.com's harnesses vs. Gemini 3.1 Pro via Google AI StudioKimi (Moonshot AI) · Bar chart · 3 points Estimated from the imageGeneral AgentsKimi (Moonshot AI) · Benchmark bar chart · 39 points Estimated from the imageCompare Categories by usecase on OpenRouterKimi (Moonshot AI) · Stacked bar chart · 10 points Estimated from the imageAI Office Bench vs General-Agent BenchKimi (Moonshot AI) · Benchmark bar chart · 6 points Estimated from the imageFigure 2: Max Logits vs Training StepsKimi (Moonshot AI) · Line chart · 20 points Estimated from the imageMMMU ProKimi (Moonshot AI) · Line chart · 100 points Estimated from the imageExecution Time to Solve Tasks of Different ComplexityKimi (Moonshot AI) · Scatter plot · 34 points Estimated from the imageMulti-Objective Performance OptimizationKimi (Moonshot AI) · Scatter plot · 10 points Estimated from the imageNet worth over timeKimi (Moonshot AI) · Line chart · 45 points Estimated from the imageAI Model Performance ComparisonKimi (Moonshot AI) · Bar chart · 14 points Estimated from the imageCodingKimi (Moonshot AI) · Benchmark bar chart · 41 points Estimated from the imageModel Comparison: Cost vs PerformanceKimi (Moonshot AI) · Line chart · 6 points Estimated from the imageKimi Claw BenchKimi (Moonshot AI) · Grouped bar chart · 2 points Estimated from the imageKimi K2.5 Agent Swarm vs Kimi K2.5 vs Claude Opus 4.5Kimi (Moonshot AI) · Benchmark bar chart · 9 points Estimated from the imageModel Performance Across BenchmarksKimi (Moonshot AI) · Benchmark bar chart · 48 points Estimated from the imageTraining Accuracy vs StepsKimi (Moonshot AI) · Line chart · 100 points Estimated from the imageOSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer EnvironmentsKimi (Moonshot AI) · Table chart · 5 points Estimated from the imageHarvey LAB-AA against hallucinations: Criterion Pass Rate vs. Material Hallucinations per Taskartificial-analysis · Scatter plot · 29 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Benchmark bar chart · 101 points Estimated from the imageOutput Tokens Used to Run Artificial Analysis Intelligence Index: Non-reasoning Modelsartificial-analysis · Bar chart · 17 points Estimated from the imageKimi K3 2.8T Agentic Revenue & Profit Estimatessemianalysis · Stacked bar chart · 16 points Estimated from the imageFew flagship releases by China's leading AI labs are open under a restrictive licenseepoch-ai · Table chart · 71 points Estimated from the imageAA-Briefcase Elo - Cost per Task (USD)artificial-analysis · Scatter plot · 38 points Estimated from the imageCode Arena: WebDevarena · Bar chart · 15 points Estimated from the imageCost (USD) to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 40 points Estimated from the imageArtificial Analysis Intelligence Index v3.0 incorporates 10 evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, AIME 2025, IFBench, AA-LCR, Terminal-Bench Hard, τ²-Bench Telecomartificial-analysis · Benchmark bar chart · 35 points Estimated from the imageVision Kimi-K2.6: Top Open Modelarena · Bar chart · 15 points Estimated from the imageWeighted cost per Intelligence Index Taskartificial-analysis · Benchmark bar chart · 88 points Estimated from the imageCost per GPU/Total Tokens (Owning - Hyperscaler) vs. Interactivitysemianalysis · Scatter plot · 36 points Estimated from the imageText Kimi-K3: Overall Ranked #9arena · Bar chart · 23 points Estimated from the imageToken Throughput per GPU vs. Interactivitysemianalysis · Line chart · 18 points Estimated from the imageCost (USD) to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 78 points Estimated from the imageCost per Intelligence Indexartificial-analysis · Benchmark bar chart · 114 points Estimated from the imageGet Data DescriptionKimi (Moonshot AI) · Table chart · 4 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 26 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Bar chart · 33 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 32 points Estimated from the imageFigure 1: Performance of Kimi-Audio and previous audio langauge models on various benchmarksKimi (Moonshot AI) · Table chart · 75 points Estimated from the imageKimi K2.5 Instant #2 open model in Vision Arenaarena · Table chart · 11 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 58 points Estimated from the imageQwen3.5-397B-A17B #top2 overall model in Visionarena · Table chart · 28 points Estimated from the imageAgent Kimi K2.7 Code Leaderboard: Ranked #19arena · Benchmark bar chart · 25 points Estimated from the imageOutput Tokens Used to Run All Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 44 points Estimated from the imageToken Efficiency before and after Toggle across BenchmarksKimi (Moonshot AI) · Table chart · 14 points Estimated from the imageLMArena.ai WebDev Arena Leaderboardarena · Table chart · 33 points Estimated from the imageOutput tokens to run all evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 87 points Estimated from the imageCost per Intelligence Index Taskartificial-analysis · Scatter plot · 31 points Estimated from the imagePerformance comparison across different vision-text joint-training strategiesKimi (Moonshot AI) · Table chart · 30 points Estimated from the imageModel Performance ComparisonKimi (Moonshot AI) · Benchmark bar chart · 25 points Estimated from the imagePass Rate on MiniF2F-testKimi (Moonshot AI) · Line chart · 57 points Estimated from the imagePerceptionBench (Ours)Kimi (Moonshot AI) · Table chart · 12 points Estimated from the imageEvaluation ResultsKimi (Moonshot AI) · Table chart · 58 points Estimated from the imageEvaluation Time: 2025-10-10Kimi (Moonshot AI) · Table chart · 107 points Estimated from the imageThe AI that actually does thingsKimi (Moonshot AI) · Stacked bar chart · 32 points Estimated from the imageVoxelBench LeaderboardKimi (Moonshot AI) · Table chart · 81 points

Z.ai 78

Publisher's numbersReplica 9.0LLM Performance EvaluationZ.ai · Benchmark bar chart · 30 points Partly publisher's numbersLLM Benchmarks: Terminal Bench 2.1, DeepSWE v1.1, Agents' Last Exam, AutomationBench v1.0.6, HLE w/ Tools, GDPVal-AA v2Z.ai · Benchmark bar chart · 33 points Estimated from the imageReplica 7.7Coding Performance SWE-Bench Pro EvaluationZ.ai · Bar chart · 7 points Estimated from the imageReplica 7.9Coding Performance EvaluationZ.ai · Bar chart · 7 points Partly publisher's numbersReplica 9.2LLM Performance Bench 3.0Z.ai · Benchmark bar chart · 47 points Partly publisher's numbers8 LLM benchmarks: Humanity's Last Exam, SWE-bench Verified, SWE-bench Multilingual, Terminal-Bench 2.0, BrowseComp, MCP-Atlas, τ²-Bench, Vending Bench 2Z.ai · Benchmark bar chart · 57 points Partly publisher's numbersReplica 8.1Model Performance ComparisonZ.ai · Benchmark bar chart · 35 points Partly publisher's numbers8 LLM benchmarks Performance: AIME 25, LiveCodeBench v6, GPQA, HLE, SWE-bench Verified, Terminal-Bench, τ²-Bench, BrowseCompZ.ai · Benchmark bar chart · 40 points Estimated from the imageCode Arena 3.7 Max (20260517) : #3arena · Bar chart · 19 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Table chart · 33 points Estimated from the imageReplica 7.6Artificial Analysis Intelligence Index v4.1artificial-analysis · Bar chart · 41 points Estimated from the imageReplica 8.3Code Arena: Frontend #2arena · Bar chart · 14 points Estimated from the imageReplica 7.7Agent Arena Leaderboardarena · Bar chart · 27 points Estimated from the imageReplica 4.8Agentic Coding Performance by Effort LevelZ.ai · Line chart · 11 points Estimated from the image函数绘图板Z.ai · Line chart · 64 points Estimated from the imageModel Performance ComparisonZ.ai · Benchmark bar chart · 10 points Estimated from the imageOpen-source LLMs BenchmarksZ.ai · Table chart · 68 points Estimated from the imageReplica 7.9Average Token Usage per InteractionZ.ai · Bar chart · 4 points Partly publisher's numbersLong-Horizon Task EvaluationZ.ai · Benchmark bar chart · 15 points Estimated from the imageZ.ai Agentic Code Bench v1.0, evaluated on Claude Code 2.1.207Z.ai · Line chart · 15 points Estimated from the imageModel PerformanceZ.ai · Bar chart · 10 points Estimated from the imageOptimization Iteration vs QPSZ.ai · Scatter plot · 7 points Estimated from the imageReplica 8.58 LLM benchmarks: AIME 25, GPQA, LiveCodeBench v6, HLE, BrowseComp, SWE-bench Verified, Terminal-Bench, τ²-BenchZ.ai · Benchmark bar chart · 44 points Estimated from the imageLLM Performance on Harvey LABZ.ai · Bar chart · 7 points Estimated from the imageCC-Bench: GLM-4.5's Experience with Agentic Coding in Real-world Development ScenariosZ.ai · Stacked bar chart · 24 points Estimated from the imageArtificial Analysis Intelligence Index v3.0Z.ai · Bar chart · 44 points Estimated from the imageReplica 7.8LLM Benchmarks: SWE-bench Verified, τ²-Bench, BrowseComp, AIME 25, GPQA, HLEZ.ai · Benchmark bar chart · 18 points Estimated from the imageAverage Tool Usage per InteractionZ.ai · Bar chart · 4 points Estimated from the imageCC-Benchmark-V1.1: GLM 4.6s Experience with Agentic Coding in Real-worldZ.ai · Benchmark bar chart · 28 points Estimated from the imageMoney Balance Over TimeZ.ai · Line chart · 56 points Estimated from the imageCC-Bench-V2: GLM-4.7 vs. GLM-5 vs. Claude Opus 4.5Z.ai · Benchmark bar chart · 15 points Estimated from the imageFrontier Models: Intelligence vs. Cost per TaskZ.ai · Scatter plot · 6 points Estimated from the imageReplica 7.0Coding EvaluationZ.ai · Bar chart · 3 points Partly publisher's numbersLLM Benchmarks: Humanity's Last Exam, SWE-bench Verified, SWE-bench Multilingual, Terminal-Bench 2.0, BrowseComp, MCP-Atlas, τ²-Bench, Vending Bench 2Z.ai · Benchmark bar chart · 51 points Partly publisher's numbers8 LLM Benchmarks: SWE-bench Pro, Terminal-Bench 2.1 (Terminus), NL2Repo, DeepSWE, ProgramBench, MCP-Atlas, Tool-Decathlon, Humanity's Last ExamZ.ai · Benchmark bar chart · 52 points Estimated from the imageAutoGLM Success Rates on Frequent TasksZ.ai · Stacked bar chart · 21 points Estimated from the imageGLM-4.7 vs GLM-4.6 in Real-world Development ScenariosZ.ai · Stacked bar chart · 9 points Estimated from the imageModel Parameters vs SWE-bench VerifiedZ.ai · Scatter plot · 9 points Estimated from the imageLLM Performance Evaluation: Agentic, Reasoning, And Coding BenchmarksZ.ai · Bar chart · 14 points Estimated from the imageAgentic Coding: SWE-Bench ProZ.ai · Bar chart · 7 points Estimated from the imageGLM-4.5: Agentic, Reasoning, and Coding (ARC)Z.ai · Benchmark bar chart · 56 points Estimated from the imageReplica 9.28 LLM benchmarks: AIME 25, GPQA, LiveCodeBench v6, HLE, BrowseComp, SWE-bench Verified, Terminal-Bench, τ²-BenchZ.ai · Benchmark bar chart · 43 points Estimated from the imageKernelBench Level 3: Agentic ML Workload OptimizationZ.ai · Line chart · 4 points Estimated from the imageModel Comparison MetricsZ.ai · Table chart · 27 points Estimated from the imageGLM-4.5's Experience with Agentic Coding in Real-world Development ScenariosZ.ai · Stacked bar chart · 9 points Estimated from the imageAccuracy ComparisonZ.ai · Bar chart · 4 points Estimated from the imageBenchmarksZ.ai · Table chart · 36 points Estimated from the imageOpen-source LLMs BenchmarksZ.ai · Table chart · 39 points Estimated from the imageCC-Bench-V1.1: GLM 4.6's Experience with Agentic Coding in Real-world Development ScenariosZ.ai · Stacked bar chart · 12 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Benchmark bar chart · 50 points Estimated from the imageAGENTIC WEBDEVarena · Table chart · 10 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Bar chart · 37 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Bar chart · 35 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Bar chart · 33 points Estimated from the imageElo rating-AA v2.1 on real-world work tasksartificial-analysis · Bar chart · 28 points Estimated from the imageCost per Intelligence Index Taskartificial-analysis · Bar chart · 25 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Grouped bar chart · 27 points Estimated from the imageArtificial Analysis Cyber Indexartificial-analysis · Benchmark bar chart · 62 points Estimated from the imageGLM-5.2 sets a new ECI record for open-weight models, but remains far behind state-of-the-art performanceepoch-ai · Benchmark bar chart · 14 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Grouped bar chart · 27 points Estimated from the imageText Arena GLM-5.2 (Max) vs GLM-5.1 Category Rankingsarena · Table chart · 28 points Estimated from the imageIntelligence vs. Cost per Intelligence Index Taskartificial-analysis · Scatter plot · 33 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Benchmark bar chart · 41 points Estimated from the imageOutput Tokens used to run evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Stacked bar chart · 82 points Estimated from the imageThird parties took over serving GLM 5.3 Flash within three weeks of its open-weight releaseepoch-ai · Stacked bar chart · 4 points Estimated from the imageArtificial Analysis Intelligence Indexartificial-analysis · Grouped bar chart · 44 points Estimated from the imageGLM-5.2 (Max): Ranked #10arena · Table chart · 12 points Estimated from the imageCode Arena: WebDev Rank Top 15arena · Bar chart · 20 points Estimated from the imageOutput Tokens used to run all evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 54 points Estimated from the imageSpeed vs. Price: GLM-4.7 Providersartificial-analysis · Scatter plot · 8 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 36 points Estimated from the imageGDPval-AA scores for agentic performance on real-world work tasks using web and shell access via Stirrup, an open-source harness developed by Artificial Analysisartificial-analysis · Bar chart · 30 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Bar chart · 36 points Estimated from the imageIntelligence vs. Cost to Run Artificial Analysis Indexartificial-analysis · Scatter plot · 20 points Estimated from the imageCode Arena: Frontend GLM-5.2 (Max): Head-to-Headarena · Benchmark bar chart · 42 points Estimated from the imagePareto Frontier for Code Arena: WebDevarena · Scatter plot · 17 points Estimated from the imageAA-Omniscence Indexartificial-analysis · Grouped bar chart · 32 points Estimated from the imagePer-Layer KV-Cache SizeZ.ai · Line chart · 24 points

Apple 33

Estimated from the imageReplica 9.0GPQA Diamond (Scientific Reasoning)artificial-analysis · Bar chart · 19 points Estimated from the imageSystem Output Throughput vs Output Speed per Queryartificial-analysis · Line chart · 95 points Estimated from the imageAccuracy / Percentile (%) by Benchmarkdeepseek · Benchmark bar chart · 30 points Estimated from the imageArtificial Analysis Intelligence Index v2.2artificial-analysis · Bar chart · 22 points Estimated from the imageArtificial Analysis Omniscence Index vs. Total Parametersartificial-analysis · Scatter plot · 44 points Estimated from the imagePerformance vs Latency for MobileCLIP and Baselinesapple · Scatter plot · 33 points Estimated from the imagedeepseek-r1-lite-preview AIME accuracy with test-time compute scalingdeepseek · Line chart · 17 points Estimated from the imageModel Accuracy vs Latency Comparisonapple · Scatter plot · 27 points Estimated from the imageAverage Score (16K max context) vs. End-to-End Generation Timeartificial-analysis · Scatter plot · 39 points Estimated from the imageGPQA Diamond vs. Inference Time; GPQA Diamond scoreartificial-analysis · Scatter plot · 15 points Estimated from the imageArtificial Analysis Intelligence Index incorporates 7 evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, AIME, MATH-500artificial-analysis · Benchmark bar chart · 21 points Estimated from the imageAgent MiniMax M3: Ranked #18arena · Bar chart · 26 points Estimated from the imageArtificial Analysis Intelligence Index incorporates 7 evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, MATH-500artificial-analysis · Bar chart · 22 points Estimated from the imageMATH Level 5 accuracy for DeepSeek-R1-0528 compared to select leading modelsepoch-ai · Benchmark bar chart · 6 points Estimated from the imageIntelligence Parameters vs. Active Parametersartificial-analysis · Scatter plot · 15 points Estimated from the imageArtificial Analysis Intelligence Index v2artificial-analysis · Bar chart · 22 points Estimated from the imageInput/Output Pricing for o1-Class Inference Models ($/1M Tokens)deepseek · Grouped bar chart · 12 points Estimated from the imageSWE-bench verified accuracy for DeepSeek-R1-0528 compared to select leading modelsepoch-ai · Benchmark bar chart · 15 points Estimated from the imagePeak Inference Memoryartificial-analysis · Bar chart · 32 points Estimated from the imageArtificial Analysis Intelligence Index incorporates 7 evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, AIME, MATH-500artificial-analysis · Bar chart · 26 points Estimated from the imageIntelligence vs. Output Tokens Used in Artificial Analysis Intelligence Indexartificial-analysis · Scatter plot · 4 points Estimated from the imageCode MiniMax M3: Frontend Ranks #7arena · Bar chart · 21 points Estimated from the imageExpert Prompts vs Open Frontier Gaparena · Grouped bar chart · 30 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 35 points Estimated from the imageThroughput per User vs Output Throughput per GPUsemianalysis · Line chart · 30 points Estimated from the imageDeepSeek R1 O528 • FP8 • 8K / 1K • Source: SemiAnalysis InferenceX™ • Updated: 02/17/2026semianalysis · Scatter plot · 19 points Estimated from the imageHealth Age Historyapple · Line chart · 5 points Estimated from the imageTCO $/GPU/Million Total Tokens (Owning) vs Interactivitysemianalysis · Line chart · 12 points Estimated from the imageModel Performance vs Latencyapple · Scatter plot · 13 points Estimated from the imageInterpolated Token Throughput per GPU (tok/s/gpu)semianalysis · Line chart · 85 points Estimated from the imageArtificial Analysis Intelligence Index incorporates 7 evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, AIME, MATH-500artificial-analysis · Bar chart · 17 points Estimated from the imageAIME 2024deepseek · Bar chart · 8 points Estimated from the imageArtificial Analysis Intelligence Index v4.0artificial-analysis · Bar chart · 32 points

MiniMax 18

Estimated from the imageModel Performance ComparisonMiniMax · Benchmark bar chart · 18 points Partly publisher's numbersReplica 8.8Model Performance Comparison Across BenchmarksMiniMax · Benchmark bar chart · 45 points Partly publisher's numbersReplica 8.1Model Benchmarks ComparisonMiniMax · Benchmark bar chart · 36 points Partly publisher's numbersReplica 8.3Model Performance Comparison Across BenchmarksMiniMax · Benchmark bar chart · 46 points Estimated from the imageReplica 9.0Model Performance ComparisonMiniMax · Benchmark bar chart · 18 points Partly publisher's numbersModel Performance Comparison Across BenchmarksMiniMax · Benchmark bar chart · 63 points Estimated from the imageReplica 7.7Model Performance by SubsetMiniMax · Benchmark bar chart · 20 points Estimated from the imageModel Performance Comparison Across BenchmarksMiniMax · Benchmark bar chart · 33 points Partly publisher's numbersModel Benchmarks ComparisonMiniMax · Benchmark bar chart · 40 points Estimated from the imageReplica 7.8Model Performance ComparisonMiniMax · Benchmark bar chart · 28 points Estimated from the imageSWE-bench Verified Score EvolutionMiniMax · Line chart · 33 points Estimated from the imageReplica 8.3Medal Rates Over TimeMiniMax · Line chart · 125 points Estimated from the imageELO scores-AA gold subset of GDPval, independently assessed by Artificial Analysisartificial-analysis · Bar chart · 32 points Estimated from the imageArtificial Analysis Intelligence Index by Open Weights / Proprietaryartificial-analysis · Stacked bar chart · 47 points Publisher's numbersReplica 6.4Model Benchmarks ComparisonMiniMax · Benchmark bar chart · 61 points Estimated from the imageModel Performance Across BenchmarksMiniMax · Benchmark bar chart · 50 points Estimated from the imageModel Performance Comparison Across SubsetsMiniMax · Benchmark bar chart · 20 points Estimated from the imageCost (USD) to Run all Artificial Analysis Artificial Analysis Intelligence Indexartificial-analysis · Stacked bar chart · 29 points

Mistral AI 31

Estimated from the imageModel Performance Comparison (Instruct)Mistral AI · Benchmark bar chart · 8 points Estimated from the imageBase Model Benchmark ComparisonMistral AI · Benchmark bar chart · 20 points Estimated from the imageScore vs. Output LengthMistral AI · Benchmark bar chart · 20 points Estimated from the imageReplica 8.3Agentic Benchmarks vs previous Mistral coding modelsMistral AI · Bar chart · 4 points Estimated from the imageMath, instruction following vs competing models*Mistral AI · Grouped bar chart · 37 points Estimated from the imageAgentic Benchmarks vs previous Mistral modelsMistral AI · Benchmark bar chart · 38 points Estimated from the imageReplica 8.1Agentic Benchmarks vs competing models*Mistral AI · Benchmark bar chart · 50 points Estimated from the imagePerformance comparison across internal modelsMistral AI · Benchmark bar chart · 50 points Estimated from the imagePerformance Comparison Across Internal ModelsMistral AI · Benchmark bar chart · 24 points Estimated from the imageScore vs. Output LengthMistral AI · Benchmark bar chart · 16 points Estimated from the imageScore vs. Output LengthMistral AI · Benchmark bar chart · 10 points Estimated from the imageCost per Intelligence Index Taskartificial-analysis · Scatter plot · 30 points Estimated from the imageMistral Large 3 is a new ranking #28 in Text model overallarena · Table chart · 18 points Estimated from the imageOutput Reasoning Models Used to Run Artificial Analysis Intelligence Indexartificial-analysis · Bar chart · 17 points Estimated from the imageMistral Large 3 is the #1 open model in Codingarena · Table chart · 36 points Estimated from the imageVals.ai - Harvey's Legal Agent BenchmarkMistral AI · Bar chart · 8 points Estimated from the imageFigure 1: Confidence Intervals on Model Strength (via Bootstrapping)arena · Bar chart · 25 points Estimated from the imagePricing per 1M Tokens; Lower and Output is better Pricesartificial-analysis · Grouped bar chart · 6 points Estimated from the imageMistral Medium 3 ranks at #11 in the Arena!arena · Table chart · 52 points Estimated from the imageModel Accuracy Comparisonartificial-analysis · Bar chart · 6 points Estimated from the imagePricing: USD per 1M Tokensartificial-analysis · Grouped bar chart · 28 points Estimated from the imageIntelligence vs. Active Parametersartificial-analysis · Scatter plot · 30 points Estimated from the imageLMArena ELO ScoreMistral AI · Bar chart · 6 points Estimated from the imagePerformance / MMLU-Proarena · Scatter plot · 4 points Estimated from the imageIntelligence Evaluationsartificial-analysis · Bar chart · 101 points Estimated from the imageOutput Tokens used to Run all Evaluations in the Artificial Analysis Intelligence Indexartificial-analysis · Benchmark bar chart · 33 points Estimated from the imageOpen Weights: Frontier Indexartificial-analysis · Line chart · 45 points Estimated from the imageArtificial Analysis Intelligence Index v3.0artificial-analysis · Bar chart · 25 points Estimated from the imageArtificial Analysis - AutomationBenchMistral AI · Bar chart · 8 points Estimated from the imageArtificial Analysis Cyber IndexMistral AI · Bar chart · 19 points Estimated from the imageMMLU vs. Priceartificial-analysis · Scatter plot · 13 points

Questions people ask

What is chart data?

For each chart a lab or tracker published, one file with every number on it, where the number came from, and how sure we are. 2391 charts and 73,917 points so far.

Are the numbers exact?

Only 2 chart is fully the publisher's own text numbers and 29 are partly. The rest are read from the image and labelled as estimates. Every page shows which is which.

Can I download the files?

Yes. Each chart page has JSON, CSV and an agent skill file behind the email gate. The API page lists the static files.