November 17, 2026

Mistral’s ‘Le Chonk’ Model Excels at Coding, Trails China Overall

Mistral has released the weights for its largest model, Large 4, which independent testing shows is fast and affordable but still lags top Chinese open models on broader benchmarks.
Mistral’s ‘Le Chonk’ Model Excels at Coding, Trails China Overall

Paris-based Mistral has published the weights for Mistral Large 4, nicknamed “Le Chonk,” on Hugging Face three weeks after its public preview, according to the-decoder.com. The model accepts text and images and outputs text and code, with roughly one trillion total parameters and 52 billion active per step. Running it requires substantial hardware, with the standard version needing eight Nvidia B300 GPUs.

Since the preview, Mistral continued training the model with reinforcement learning, which the company says produced large gains in coding and cyber defense. On Terminal-Bench 4.0, Large 4’s score rose from 28 to 41 percent, edging past GLM-5.3 at 40 percent. Mistral says Large 4 scores 48 percent on Artificial Analysis’ Coding Agent Index, roughly matching GLM-5.3 while using 47 percent fewer tokens. Claude Opus 5.5 tops that index at 62 points.

The model trails Chinese rivals more broadly, however. The-decoder.com reports that in Artificial Analysis’ Intelligence Index, Large 4 scores 26 points, behind DeepSeek V4.1 Flash and GLM-5.3 at 32 each, and MiMo-V2.6-Pro at 30. Moonshot’s unreleased Kimi K3.1 Pro leads Terminal-Bench 4.0 at 67 percent and is expected to widen the gap once its weights ship by November 2026.

Mistral calls Large 4 the leading open model for cyber defense, citing a 51 percent score on Artificial Analysis’ Cyber Index, though two open models rank higher. The company says Large 4 rejects 96.6 percent of malicious cyber requests, the highest among open models, while acknowledging it could still aid real attacks and promising tighter safeguards.

Large 4’s main advantages are speed and price, generating 161 tokens per second and costing $2.06 per average index task, compared with $5.54 for the preview. It is released under a modified MIT license, positioned by Mistral as sovereign, Europe-trained AI.

Based on reporting by the-decoder.com.

Leave a Reply

Your email address will not be published. Required fields are marked *