Mistral ships 1T-param Le Chonk, tops Surge human coding eval among open-weight models

_AndrewZhao · x · 2026-10-07

Mistral released Mistral Large 4, aka Le Chonk: 1T total parameters with 49B active, natively multimodal, and claimed the best open-weights model from the US or Europe on aggregated benchmarks. Mistral says it's SOTA on cyber defense, manufacturing, and finance workloads, surpasses closed frontier models on visual grounding, and is deployable from Europe via its own cloud. API is live today; open weights due end of October.

For the launch, Mistral commissioned a blind human eval by Surge's expert software engineers across five frontier models: Le Chonk ranked #1 among open-weight models and #2 overall, behind only Opus 5. Surge's pitch: unit tests ask "does it work?" — human code review asks "would you merge it?", and professional taste is what makes code shippable.

Related event: Mistral Launches Mistral Large 4, a 1T-Parameter Open-Weight Flagship(86 posts)→

Original post →

More from Models

Models channel →