Mistral Large 4 enters Code Arena: WebDev leaderboard at 45th place

Mistral Large 4 debuted on Arena.ai's Code Arena: WebDev benchmark at 45th place with an Elo rating of 1,534 points. The evaluation represents a 304-point gain over Mistral Large 3 at 130th place and a 271-point increase over Mistral Medium 3.5 at 122nd place.

The model performed within two points of Claude Opus 4.8 (High) while running at a blended rate of $3.47 per million tokens versus $20 per million tokens for Opus. Arena.ai noted that other higher-performing models remain available for less, keeping Mistral Large 4 just off the benchmark's Pareto frontier. Mistral AI plans to release open weights for the model at the end of October.