Mistral opens Large 4 weights, leads open-weight coding
Mistral has released the weights of Mistral Large 4 (Le Chonk), its 1-trillion-parameter natively multimodal model, after its public preview earlier this month. It ranks as the best open-weight model for coding and among the world's most capable open models overall.
For teams that want to self-host coding agents or security tooling, this means a capable model they can inspect, customize, and run on their own infrastructure.
Key Features at a Glance
- Coding efficiency: 48% on the Artificial Analysis Coding Agent Index, effectively tied with GLM-5.3, while using 47% fewer tokens per task on average – cheaper agent runs at scale.
- Terminal-Bench 4.0: 41%, up 13 percentage points from the preview, with 30% fewer output tokens – stronger terminal-based coding workflows.
- Three reasoning modes: Low, Medium, and High (default) – lets developers trade cost against precision for each task.
- Cyber Index: 51% on the Artificial Analysis Cyber Index, plus an average refusal rate of 96.6% on malicious cyber requests – helpful for defensive work without unnecessary refusals.
- Agent security: Resists 97.1% of attacks on the B3 Agent Security Benchmark, up from 93.3% in the preview – safer agents without extra tuning.
- Finance and browsing: 73.3% on Finch (second only to Kimi K3 among open models) and 84.5% on BrowseComp – useful for spreadsheet and web-research workflows.
- Sovereignty: Trained from scratch on European infrastructure, with inspectable, self-hostable weights.
Availability
Mistral Large 4 is available today to try, deploy, and customize through Mistral Studio, and as a managed service through Microsoft Foundry, with private deployment options for Azure developers.
