DeepSeek-V4-Pro-0813: benchmarks, price and rank
DeepSeek-V4-Pro-0813 from DeepSeek is 37th of 123 rated AI models on Spezia's leaderboard, strongest on Humanity's Last Exam and Toolathlon. It costs $1.30 in, $2.60 out per million tokens.
- 37th of 123 rated models
- Rating 1,426
- Rated on 11 benchmarks
- Open weights
- Released Apr 24, 2026
- 1 million tokens of context
Where it is strongest
- 6th of 74 on Humanity's Last Exam
- 3rd of 37 on Toolathlon
- 3rd of 15 on CyberGym
Where it falls behind
- 47th of 60 on Vending-Bench 2
- 37th of 56 on APEX-Agents
- 40th of 66 on FrontierMath Tier 4
Price: $1.30 in, $2.60 out per million tokens.
Its headline results
The benchmarks that count most in the rating, with its place among the models measured the same way, and where each number came from.
- APEX-Agents47.3%
Long tasks written by investment banking analysts, management consultants and corporate lawyers from their own work, done across files and apps in a realistic work setup.
37th of 56. Independent result from Epoch AI.
- Humanity's Last Exam60%
About 2,500 questions written by experts in more than a hundred fields to be as hard as they could make them, many needing a diagram or image.
6th of 74. The maker's own number, from DeepSeek, via LLM Stats.
- Vending-Bench 2$3,285
The model runs a simulated vending machine business for a year on its own: finding suppliers, ordering stock, setting prices and paying the daily fees.
47th of 60. Independent result from Epoch AI.
- Toolathlon74.1%
Long tasks across many real apps and their tools, such as email, calendars, spreadsheets, code repositories and online stores.
3rd of 37. The maker's own number, from DeepSeek, via LLM Stats.
- SimpleQA Verified52.9%
Short factual questions with one checked answer, asked without web search.
27th of 78. Independent result from Epoch AI.
- FrontierMath64.6%
Hundreds of new, unpublished math problems written by mathematicians, from hard undergraduate level up to research, with answers a computer can check.
39th of 76. Independent result from Epoch AI.
- LMArena text1,451
People put a question to two anonymous models and vote for the better answer.
38th of 107. Independent result from LMArena.
- LMArena WebDev1,581
People ask two anonymous models to build the same web page or app and vote for the better one.
34th of 98. Independent result from LMArena.
- CyberGym83.3%
Security tasks on real software: finding vulnerabilities and showing they can be triggered.
3rd of 15. The maker's own number, from DeepSeek, via LLM Stats.
- SWE-bench Verified77.6%
Real bug reports and feature requests from popular open-source Python projects. The model has to change the code so the project's hidden tests pass.
Independent result from Epoch AI.
4 more results in its full profile.
Every result with its source, what it costs per task, what jumped with its release, side by side with any other model, and answers to your questions about when to use it.
Models ranked near it
How the rating works, and every benchmark it uses: the model leaderboard.