DeepSeek V3 benchmarks and performance, in charts
Every chart we tracked that shows DeepSeek V3, from the lab itself and from the trackers people trust, with how far each one travelled. Includes DeepSeek V3.1, DeepSeek V3.2.
DeepSeek V3 on its own benchmark table
Drawn from the numbers huggingface.co published as text (Standard Benchmarks, read Mar 27, 2025). Each panel is one test; the highlighted bar is DeepSeek V3. Source: huggingface.co.
See the numbers
| English | Qwen2.5 72B 0.638, DeepSeek-V2 0.606, DeepSeek-V3 0.548, LLaMA3.1 405B 0.542 |
|---|---|
| Code | DeepSeek-V3 65.2, LLaMA3.1 405B 54.9, Qwen2.5 72B 53, DeepSeek-V2 43.3, # Shots 0 |
| Math | DeepSeek-V3 89.3, Qwen2.5 72B 88.3, LLaMA3.1 405B 83.5, DeepSeek-V2 81.6, # Shots 8 |
| Chinese | LLaMA3.1 405B 83, DeepSeek-V3 82.7, Qwen2.5 72B 82.5, DeepSeek-V2 82, # Shots 5 |
| Multilingual | DeepSeek-V3 79.4, Qwen2.5 72B 74.8, LLaMA3.1 405B 73.8, DeepSeek-V2 64, # Shots 5 |
Official charts from DeepSeek
As published on X, Hugging Face, GitHub and their blog. Each card opens our own page for the chart.
From DeepSeek, measured on X
From benchmark trackers and others
Free with your email
See all 75 DeepSeek V3 charts
Every official launch chart and independent benchmark we found for DeepSeek V3, with who posted each one. Plus Monday's email: the week's charts that beat their usual likes.
Unlocks all 35 giants. One email a week, leave any time. What you get · Log in
From benchmark trackers and others more
More charts that feature DeepSeek V3
Benchmark trackers, other labs and researchers. Each card opens our own page for the chart.
%201_2flnuY.webp?dpl=6accf0d261fecbe6ba702f4d)
Make it yours
Your numbers, in a look-alike of the original.
Starts from the real numbers of "Accuracy Percentile by Model and Benchmark" by DeepSeek. Replace every number with yours before you post. Make one like this, step by step
Look-alike score 8.2 of 10. Free pictures carry a small "Made with ViralChart" line. That line is how other people find us.
Questions people ask
How does DeepSeek V3 perform?
We hold 17 charts that show DeepSeek V3 (including DeepSeek V3.1, DeepSeek V3.2), posted between Dec 2024 and Jul 2026, 1 of them from DeepSeek. The most seen is "LLM benchmark comparison" by DeepSeek with 7,427,309 views.
DeepSeek V3 vs Qwen3: which is better?
3 charts put them side by side. Which wins depends on the benchmark; each chart shows its own test and who published it.
Who publishes DeepSeek V3 benchmark charts?
DeepSeek, Qwen, Artificial Analysis, Arena, NVIDIA, SemiAnalysis, @ArtificialAnlys, @MiniMax_AI, @Alibaba_Qwen, @Zai_org