1. Models
  2. Meta
  3. GPQA Accuracies for Leading Models
E
@EpochAIResearchX · Jul 31, 2024

GPQA Accuracies for Leading Models

@EpochAIResearch's dumbbell chart drew 6.9K views on X: 1.3x @epochairesearch's median post and #879 of 1,162 AI chart posts we hold by views. Read Oct 12, 2026.

GPQA accuracy comparison for leading AI models

XDataDumbbell chartLlama 3All Meta modelsGPT-4oClaude 3.5 SonnetLlama 3.1 405B InstructMistral Large 2407
Published by @EpochAIResearch on X, Jul 31, 2024. Source: x.com
47/ 100

ViralChart score

Small but loved

Fewer views than the median post but a higher engagement rate: a smaller crowd that acted on it.

#667 of 1,162 AI chart posts we hold · #21 of 59 @epochairesearch posts. Score = views 40%, likes 30%, engagement rate 15%, save rate 15%, each as a percentile of every AI chart post we hold.

Our read

It sits in the Small but loved quadrant, at #667 of 1,162 AI chart posts.

Its 0.27% save rate sits at the 96th percentile among AI chart posts, against a 0.076% median.

At 6.9K views it runs 1.3x @EpochAIResearch's 5.3K company median, #24 of 59 @EpochAIResearch posts.

The 47 score lands at #667 of 1,162 AI chart posts, below the field.

top 4%save rate

0.274% of viewers bookmarked it to come back to. The median post gets 0.076%.

top 13%engagement rate

1.31% of viewers liked, reposted, quoted or replied. The median post gets 0.73%.

0.5xthe median dumbbell chart post

Across 8 dumbbell chart posts we hold, the median gets 15.2K views.

1.3x@epochairesearch's median post

6.9K views, where @epochairesearch's typical post gets 5.3K.

Public X numbers from our read on Oct 12, 2026. Peers: 1,162 AI chart posts we hold and 59 from @epochairesearch.

Where it sits

Reach against resonance

Each dot is a post. Right is more views, up is a higher engagement rate. The lines are the medians, so the four corners are the four kinds of post.

Among 1,162 AI chart posts we hold

Score #667 of 1,162

Views6.9K#879bottom 24%
Likes76#802bottom 31%
Engagement rate1.31%#147top 13%
Save rate0.274%#49top 4%

Among 59 @epochairesearch posts

Score #21 of 59

Views6.9K#24top 40%
Likes76#27top 45%
Engagement rate1.31%#31bottom 48%
Save rate0.274%#7top 11%

Growth

One read so far

Views6.9K
Age at our read803 days
Views per day of life9views divided by days since it was posted

Dots are our reads. The dotted line from the day it was posted to our first read is a gap, not data.

@epochairesearch's posts over time

Every post from this account we hold, by date and views. This one is marked; the line is the running median of the last 10.

News · Jul 31, 2024

@EpochAIResearch posts GPQA accuracy dumbbell chart for leading models

A dumbbell chart of GPQA accuracy for leading models drew 6.9K views and a 0.27% save rate, the 96th percentile among AI chart posts.

What happened

@EpochAIResearch published a dumbbell chart on GPQA accuracies for leading models on July 31, 2024. It was the account's only post that day.

The numbers

It reached 6.9K views, #879 of 1,162 AI chart posts, and #24 of 59 @EpochAIResearch posts. Its 47 score ranks #667 of 1,162 AI chart posts and #21 of 59 @EpochAIResearch posts.

The lesson

A 0.27% save rate, the 96th percentile among AI chart posts, shows readers bookmark this dumbbell chart at a rate far above the 0.076% median. The 1.31% engagement rate sits with that save behavior.

What it means

At #24 of 59 @EpochAIResearch posts by views, this chart beats the account's 5.3K median but trails the 39.9K median among all AI chart posts. The 0.27% save rate against a 0.076% median marks it as a keeper despite the reach gap.

Written by ViralChart from public numbers read on Oct 12, 2026. Every number is checked against our data before it is shown.

Updates: what @epochairesearch posted next

  1. GPQA Diamond accuracy comparison376 views
  2. MATH Level 5 pass@1 comparison for Mistral Small366 views
  3. Total Training FLOPs Over Time2.5K views
  4. Training compute of notable models369 views

Data and API

8 points read off this chart

Estimated from the image 66%
SeriesGroupValueRead
Simple-evals prompt,T=0.7GPT-4o48.5 %estimated
Llama 3.1 evals prompt,T=0GPT-4o50 %estimated
Simple-evals prompt,T=0.7Claude 3.5 Sonnet55.5 %estimated
Llama 3.1 evals prompt,T=0Claude 3.5 Sonnet50.25 %estimated
Simple-evals prompt,T=0.7Llama 3.1 405B Instruct46 %estimated
Llama 3.1 evals prompt,T=0Llama 3.1 405B Instruct49 %estimated
Simple-evals prompt,T=0.7Mistral Large 240747.5 %estimated
Llama 3.1 evals prompt,T=0Mistral Large 240746 %estimated

Max plan · API

GET https://viral-chart.com/api/posts/fb05ac8abd1c.json
GET https://viral-chart.com/api/charts/fb05ac8abd1c.json
GET https://viral-chart.com/api/charts/fb05ac8abd1c.skill.md
Make one like thisOpen the post JSONChart data JSONSkill fileAPI docs

Static files in this prototype. The post JSON holds the public numbers, ranks, score, quadrant, our reads and, when checked, our copy.

API · Max

Use this chart via API

On the Max plan you can pull this chart's numbers with the provenance of each one, as JSON or CSV, with its redraw spec, or call it from an agent over MCP. Chart id fb05ac8abd1c.

See the Max planAPI docs
PathReturnsPlan
GET/api/v1/charts/fb05ac8abd1cTitle, publisher, date, form, point count; with a Max key, the data linkOpen
GET/api/v1/charts/fb05ac8abd1c/dataEvery point with its provenance and confidence, plus the redraw spec. Add ?format=csv for CSVMax
GET/api/v1/charts/fb05ac8abd1c/render?view=1The draw-only view this page usesOpen
POST/mcpTool get_chart_data with this id, for Claude, Cursor and other agentsMax
curlMax
curl -s https://viral-chart.com/api/v1/charts/fb05ac8abd1c/data \
  -H "Authorization: Bearer $VIRALCHART_KEY"
Claude CodeMax
claude mcp add --transport http viralchart https://viral-chart.com/mcp \
  --header "Authorization: Bearer $VIRALCHART_KEY"

60 requests a minute per key; over that you get 429 with Retry-After. Up to 5 keys per account, made on your account page and shown once. Without Max, the data route answers 402 with a link to the plans. The numbers describe a chart its publisher owns, and every file carries a license note.

More charts about Llama 3

Llama 3: Llama 3.1 405B Human Evaluation
OfficialX · MetaJul 23, 2024
Llama 3.1 405B Human Evaluation50.7K views · 353 likes
Data
Llama 3: Llama context window comparison
OfficialX · MetaJul 23, 2024
Llama context window comparison10.2K views · 163 likes
Data
Llama 3: Output Tokens per Second: Llama 3.1 8B Providers
X · @ArtificialAnlysApr 10, 2025
Output Tokens per Second: Llama 3.1 8B Providers12.6K views · 176 likes
Data

More from Meta

Meta table chart: AI model benchmark comparison
Meta·Sep 2026
AI model benchmark comparison
3.8M views19.9K likes0.6% eng.
Data
Meta line chart: SAM 3.1 vs SAM 3 Performance17x
Meta·Mar 2026
SAM 3.1 vs SAM 3 Performance
344.6K views2.2K likes0.8% eng.
Data
Meta annotated chart: APhO 2026 AI Model Results3.0x
Meta·Jul 2026
APhO 2026 AI Model Results
196.6K views404 likes0.3% eng.
Data
Llama 4: LM Arena ELO score vs. cost
OfficialX · MetaApr 5, 2025
LM Arena ELO score vs. cost1.5M views · 2.3K likes
Data
Muse Glimmer: AI Model Benchmark Comparison
OfficialX · MetaAug 10, 2026
AI Model Benchmark Comparison1.4M views · 8.2K likes
Data
Muse Spark: AI Model Benchmark Comparison
OfficialX · MetaSep 2, 2026
AI Model Benchmark Comparison435.2K views · 3.1K likes
Data
Meta AI app: Hiera Model Performance Comparison
OfficialX · MetaJun 5, 2023
Hiera Model Performance Comparison173.2K views · 567 likes
Data
Meta speech models: Diariazation Error Rate Comparison
OfficialX · MetaSep 1, 2026
Diariazation Error Rate Comparison142.8K views · 1.5K likes
Data
Meta speech models: AA-WER Streaming Index - Final Transcription
OfficialX · MetaSep 1, 2026
AA-WER Streaming Index - Final Transcription142.8K views · 1.5K likes
Data

Questions people ask

Where does this chart come from?

@EpochAIResearch published it on Jul 31, 2024 on X (x.com). We keep it here, with the publisher and date credited as text, so you can study it without leaving.

Can I make a chart like this?

Use the format, not the picture. Open its data page and use Make one like this: it starts from this chart's numbers and takes yours.