Mistral Large 4 weights are out: a 1T-parameter open model tuned harder for coding and cyber defense
What is it?
Mistral Large 4 is a 1-trillion-parameter mixture-of-experts model (only some experts fire for each token) that handles both images and text. Going by the repo name, about 52B parameters are active per token. After the Oct 6 preview, Mistral ran three more weeks of reinforcement learning and has now released the improved weights in FP8. You can pick Low, Medium or High (the default) reasoning, and the repo is tagged for vLLM, SGLang and llama.cpp.
Why does it matter?
Mistral says it leads open-weight models on Cognition's private FrontierCode 2.0 at 19.7%Vendor claim and on Terminal-Bench 4.0 at 41%Vendor claim. On the Artificial Analysis Coding Agent Index it reports 48%, about level with GLM-5.3 while using 47% fewer tokens per taskVendor claim. It also says the model takes on vulnerability finding and patching without needless refusals, yet refuses 96.6% of malicious cyber requests on averageVendor claim. For teams that can't send code off-premises, that's one more serious option for a self-hosted coding or security agent.
Maturity
GAAnyone can download the weights and the model is on Mistral's API, so we're calling it GA. The license isn't plain MIT; it's Mistral's own terms (listed as license: other), so read the LICENSE file before building a commercial service on it.
How to track it
- Independent numbers: see whether Artificial Analysis posts its own scores for the final release.
- License and pricing: check the LICENSE file on Hugging Face and the model page in Mistral's docs for commercial terms and API rates.
- Security variant: watch whether the security-tuned variant mentioned in the post opens up beyond design partners.