Mistral AI Releases Flagship Model Mistral Large 4 (Le Chonk)

MonitoringAI & TechProduct releaseFirst tracked 2026-10-06Last changed 2026-10-06

Current outcome

Mistral AI launched Mistral Large 4, a 1-trillion-parameter model activating 49 billion parameters per query at $1.36/$4.18 per million input/output tokens, which it says will become the strongest open-weight model yet with weights due by end of October.

Progress timeline

2 material updates

#0110.06
Initial2026-10-06 13:15 · publication time
Mistral Large 4
Mistral released Large 4, its new flagship model, featuring strong vision and cybersecurity benchmark results and a reasoning setting limited to 'none' or 'high'.
Source evidence: Philpax · j-bu · Decrypt
#0210.06
Confirmation2026-10-06 14:31 · publication time
Mistral Large 4发布：1.05万亿参数，自称做出欧美最强开源模型
Added release details: Mistral Large 4 is a 1.05T-parameter MoE model activating 49B parameters per inference, natively supports image-text input, has up to 1M-token context; API preview is available, open weights are planned for October 27, and self-reported benchmark scores were provided.
State after update: Mistral has released flagship model Mistral Large 4 (internally nicknamed Le Chonk), an MoE model with 1.05T total parameters and 49B active parameters, multimodal support and 1M-token context; API preview is available, open weights are planned for October 27, and Mistral claims it is the strongest open-weights model from the US or Europe with self-reported benchmarks.
Source evidence: theblockbeats · TechCrunch AI

All events · Back to the feed