Mistral releases open weights for Large 4
Improved Le Chonk weights are out: a 1-trillion-parameter multimodal model. The API price is $0.70 in and $2.10 out per million tokens.
Official source (opens in a new tab)
Mistral's news post on 29 October 2026 releases the weights of an improved Mistral Large 4, which it calls Le Chonk. The model was in public preview earlier in the month. Mistral describes it as a 1-trillion-parameter, natively multimodal model, and says further reinforcement learning improved coding and defensive cybersecurity.
Mistral's figures, from that post: 19.7% on Cognition's private FrontierCode 2.0, which it says leads open-weight models. On the Artificial Analysis Coding Agent Index it scores 48%, which Mistral calls a tie with GLM-5.3 at 47% fewer tokens per task. On Terminal-Bench 4.0 it reports 41%, up 13 points from the public preview, with 30% fewer output tokens, and a lead among open-weight models. Three reasoning modes are Low, Medium, and High, and High is the default.
On the Artificial Analysis Cyber Index, Mistral reports 51%. It says the model refuses 96.6% of malicious cyber requests on average across JailbreakBench, StrongREJECT, and AgentHarm, while still attempting defensive work. It also reports 73.3% on Finch, 84.5% on BrowseComp, and 97.1% attack resistance on the B3 benchmark, up from 93.3% in preview. A specialized cybersecurity variant is limited to vetted enterprise design partners.
The weights are under the Mistral Ecosystem Open-Weight license, which Mistral describes as a modified MIT license. Use, modification, fine-tuning, and commercial use are allowed, except that a Model-as-a-Service business needs a separate commercial license from Mistral. The API price starting with this release is $0.70 per million input tokens, $0.07 per million cached input tokens, and $2.10 per million output tokens. Mistral says you can use it in Mistral Studio, on the Mistral API, or on OpenRouter, download the weights and serve them with vLLM or SGLang, or use an NVIDIA NIM or Microsoft Foundry.
Who it's for: a team that wants an open-weight model for coding or defensive security work, including a deployment it hosts itself.
Fits: Best for Vibe Coding stays Cursor, for an agent in the editor. Large 4 is a model you call or host, not that editor. The coding guide, the Cursor alternatives, and the Claude Code alternatives are unchanged. The verdict is unchanged.