# Mistral releases Large 4 weights, a coding model that trails Chinese open models overall

published: 2026-10-29T22:05:28.252Z
topic: infrastructure
source: https://the-decoder.com/paris-based-mistral-releases-le-chonk-a-strong-coding-model-that-still-trails-china-overall/
url: https://aiagentnews.io/stories/mistral-releases-large-4-weights-a-coding-model-that-trails-chinese-open-models

Mistral has released the weights for Mistral Large 4, its largest model, on Hugging Face under a modified MIT license. Independent testing finds it strong on coding, speed and cost, but behind leading open Chinese models on overall intelligence.

Mistral has published the weights for Mistral Large 4, nicknamed "Le Chonk," on Hugging Face. The model accepts text and images, has about one trillion parameters with 52 billion active per step, and can be self-hosted. Companies that want to offer it as a commercial cloud service need a separate license from Mistral.

Artificial Analysis scores the final version at 26 points on its Intelligence Index, behind DeepSeek V4.1 Flash and GLM-5.3, which score 32 each. The model's strengths are speed, at 161 tokens per second, and cost per task, which averages $2.06 but is still higher than DeepSeek V4.1 Flash at $0.60.

The API list price is $0.70 per million input tokens and $2.10 per million output tokens, down from $1.36 and $4.18 during the preview. Mistral says the model refuses 96.6 percent of malicious cyber requests, and a special penetration-testing version is being tested by select partners under review and contract restrictions.

## Why it matters if you build agents

Developers building coding agents get a cheaper, faster open-weight option they can run on their own hardware or call through the API. Per-task cost and overall benchmark standing still favor several Chinese open models.

Teams that plan to resell Large 4 as a hosted service will need a separate commercial license from Mistral. Access to the cyber-defense variant is limited to reviewed partners.