Mistral says it has released the open weights of Le Chonk, its improved Mistral Large 4 model, according to @MistralAI. The company's release page gives the API price, licence and benchmark results. The model was previewed through Mistral's API on October 6, and the weights arrived earlier than the "by the end of October" timing the company had given.

What it is and where to get it

Mistral Large 4 is Mistral's open-weight model aimed at agentic coding and cybersecurity work. The weights can be downloaded from Hugging Face for self-hosting with vLLM or SGLang. The model is also offered through Mistral Studio, the Mistral API, OpenRouter, NVIDIA-hosted endpoints and an NVIDIA NIM container.

Microsoft says it is the first cloud provider to offer Mistral Large 4, through Microsoft Foundry. The Next Web reports that the model is also available to some Vibe users.

Price

Starting today, the API price is $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens. The table compares that with the preview pricing that Mistral's release page describes.

Per 1M tokens

Preview list

Launch sale

Now

Input

$1.36

$0.68

$0.70

Output

$4.18

$2.09

$2.10

Licence

Mistral calls the weights the Mistral Ecosystem Open-Weight license, a modified MIT license. It allows free use, modification and commercialisation. The one condition is that businesses running a Model-as-a-Service offering need a separate commercial licence from Mistral.

The model offers Low, Medium and High reasoning modes, with High as the default.

Benchmarks from Mistral's charts

These figures come from Mistral's own charts and from third-party benchmarks it cites.

  • FrontierCode: Mistral reports a 19.7% score and says the model ranks first among open-weight models. The score comes from Cognition's private FrontierCode 2.0 benchmark, which judges whether a repository maintainer would merge the agent's change.

  • Artificial Analysis Coding Agent Index: 48%, effectively tied with GLM-5.3, while using 47% fewer tokens per task on average.

  • Terminal-Bench 4.0: 41%, up 13 points from the public preview, with 30% fewer output tokens.

  • Artificial Analysis Cyber Index: 51%, covering finding, reproducing and patching vulnerabilities.

Image shared by @MistralAI on X
Image shared by @MistralAI on X

Image credit: @MistralAI on X

Our read: the efficiency numbers, fewer tokens per task, matter most to anyone paying per token.

Cyber use and guardrails

Mistral reports a 96.6% average refusal rate on malicious cyber requests, which it says is the highest it measures among open models. It says the preview refused black-box attacks against live targets but still showed potential for real-attack uplift, so it is modifying its guardrails.

The release page quotes cyber partners including CrowdStrike, Cloudflare, Rubrik, Zscaler, Booz Allen, Elastic, Fortinet, ENISA and Orange Cyberdefense. A specialised cybersecurity variant is in preview only for enterprise design partners such as Orange Cyberdefense, subject to vetting and contractual safeguards.

For the API preview and the earlier timeline, see our earlier report on the Mistral Large 4 API and open-weights timeline.

Sources