Mistral Large 4: benchmarks, price and rank

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.

• Terminal-Bench 4.041%

  Hard jobs done in a real computer terminal: building and fixing software, setting up systems, and science and engineering problems.

  11th of 17. The maker's own number, from Mistral AI, via LLM Stats.

• BrowseComp84.5%

  Questions whose answers are hidden deep on the web and take many searches and pages to track down.

  13th of 56. The maker's own number, from Mistral AI, via LLM Stats.

• DeepSWE 1.161.7%

  Software engineering tasks run in the same simple agent setup for every model.

  21st of 27. The maker's own number, from Mistral AI, via LLM Stats.

• LMArena text1,414

  People put a question to two anonymous models and vote for the better answer.

  74th of 107. Independent result from LMArena.

• LMArena WebDev1,564

  People ask two anonymous models to build the same web page or app and vote for the better one.

  36th of 98. Independent result from LMArena.

• SciCode91.8%

  Research coding problems written by scientists across 16 fields, from physics and chemistry to biology and materials science.

  1st of 24. The maker's own number, from Mistral AI, via LLM Stats.

• Finance Agent v254.7%

  Real financial analysis work: finding figures in company filings and documents, multi-step calculations and written analysis.

  4th of 23. The maker's own number, from Mistral AI, via LLM Stats.

• CyberGym82%

  Security tasks on real software: finding vulnerabilities and showing they can be triggered.

  5th of 15. The maker's own number, from Mistral AI, via LLM Stats.

• Epoch Capabilities Index154.6

  One score built from dozens of benchmarks, set on a common scale so that models tested on different benchmarks can still be compared.

  46th of 94. Independent result from Epoch AI.