Introducing Mistral Large 4
- Source
- mistral.ai
- Date
Mistral Large 4 is a natively multimodal mixture-of-experts model with 1T total parameters and 49B active. The API is open in preview now, with open weights promised for the end of October and EU-hosted deployment on Mistral Cloud. Mistral claims state-of-the-art open-model results on coding, agentic, cyber and finance work, and says the model solved 18 of 19 CTF challenges and identified Cobalt Strike from an unknown binary. Those security results are vendor demos. Artificial Analysis supplied the first independent numbers: 38 on its Intelligence Index, level with GPT-6 Luna and DeepSeek V4.1 Flash, and 50 on its Cyber Index with 82% on CyberGym-E2E. It also measured cost per task at more than four times that of comparable open-weight models. Teams that need an EU-deployable model can test the preview now, but the cost gap matters for agent workloads that run many steps.

Context
Mistral AI introduced Mistral Large 4 on October 6, 2026 as a public preview: a natively with 1 trillion total parameters, 49 billion of them active for any given input, available through a preview API. Mistral's post says will follow by the end of October.
The performance claims are Mistral's, mostly from third-party benchmarks it chose to report, plus a blind human coding evaluation that ranked the model second of five, behind Claude Opus 5. The post says the model is still being refined and promises architecture and further benchmark details as the weights approach.
What is established now is the preview and the company's own results; the open weights are a stated plan.
- Mistral says ML4 was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own European datacenters, the preview runs on that same infrastructure, and a European deployment will operate under European law.
- Until the weights ship, Mistral is red-teaming ML4 with cybersecurity leaders, vetted partners and state authorities, who get the same model with reduced moderation and expanded cyber capabilities.
- Mistral reports ML4 scores 82% on a test that has models reproduce and then patch a real vulnerability, and says several closed models score near zero there because they refuse the task.
- For coding, Mistral cites 61.7% on DeepSWE v1.1, 59.4% on SWE-Atlas-QnA and 28.3% on Terminal-Bench 4, and a blind Surge AI human rating that placed the preview second of five (3.74), behind only Claude Opus 5 (4.22).
- Mistral says architecture details, more benchmarks and its post-training method will be shared as the weights approach, so for now the results are vendor-reported, with some coding scores drawn from Artificial Analysis.
- Multimodality — The ability to understand or generate more than one kind of media.
- Open weights — A model whose learned parameter values have been made available to download, under terms the publisher sets.
- multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
- mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
Checking sign-in…
Loading comments…


