Mistral Releases Mistral Large 4 Preview

Mistral has launched a public preview of Mistral Large 4, a natively multimodal model nicknamed "le Chonk." The model has 1 trillion parameters with 52 billion active parameters. Mistral trained the model from scratch on 3,800 NVIDIA Grace Blackwell GPUs. The training dataset spanned more than 160 languages, including every official language of the European Union. It is built to handle coding, agentic workflows, and multimodal tasks, with a focus on cybersecurity, finance, and legal workloads. Mistral says it will release the weights by the end of the month (October 2026), after red-teaming the model with cybersecurity partners and state authorities.

The preview API is available on Mistral Studio, priced at $1.36 per million input tokens and $4.18 per million output tokens.

The model achieves several benchmark results across multiple domains. In cybersecurity, ML4 solves 93% of challenges on Cybench, which consists of 40 exercises. It scores 82% on a test requiring the model to reproduce and patch a real vulnerability in open-source software.

On coding and terminal tasks, ML4 scores 61.7% on DeepSWE v1.1, 59.4% on SWE-Atlas-QnA, and 28.3% on Terminal-Bench 4. These results combine for a Coding Agent Index score of 49.8%. In a blind human evaluation of coding quality conducted with Surge AI using a 1–5 scale, professional annotators gave ML4 Preview a score of 3.74, placing it behind Claude Opus 5 at 4.22 but ahead of GLM-5.3 at 3.60, Kimi K3 at 3.59, and GLM-5.2 at 3.40.

For agentic workflows, the model scores 59.9% on AutomationBench, which covers 657 business workflows. In multimodal testing, ML4 scores 42% on Dense 200, compared to 41% for GPT-6-Astra.