What changed
Mistral AI announced Mistral Large 4 (ML4) on October 6, 2026 as a public preview. It is a natively multimodal mixture-of-experts model with about 1 trillion total parameters (1.05T in the documentation), 52 billion active per token and a 1.6B vision encoder. Mistral calls it its largest and most capable model to date.
The documentation lists it as mistral-large-4, version 26.10, with support for structured outputs, function calling, document Q&A, batching, agents and built-in tools. Mistral lists a 1M-token context; some third-party evaluators, including Artificial Analysis and Vals AI, list 512k.
Who can use it and when
The preview API is available now through Mistral Studio. Two weeks into the preview, the weights and their licence have still not been published, though Mistral promised them by the end of October; until then it is red-teaming the model with security firms, vetted partners and state authorities. On October 21 the documentation still linked the Hugging Face repository as 'Coming soon'.
Mistral says the preview is served from its own data centres in Europe, and that the model will be offered in several regions, including a European deployment it operates under European law.
Why it matters
Mistral positions ML4 as the strongest open-weight model built in the US or Europe, with a focus on coding agents, cybersecurity, finance and legal work. It reports 61.7% on DeepSWE v1.1, 59.4% on SWE-Atlas-QnA and 28.3% on Terminal-Bench 4, and 93% on the Cybench security challenges. It does not lead every open model: its own DeepSWE chart shows Kimi K3 higher, and on the Artificial Analysis Intelligence Index, MiMo-V2.6-Pro, GLM-5.3 and Kimi K3 score above the preview, according to The Frontier.
For teams that want a capable model they can eventually run on their own infrastructure, or that need EU-hosted inference, this is one of the more relevant releases of the month. Several results come from third-party evaluators such as Artificial Analysis and Vals AI, but they were selected and published by Mistral.
Access and limitations
Preview pricing is $1.36 per million input tokens, $0.14 per million cached input tokens and $4.18 per million output tokens. A 50% launch discount announced for the first two weeks no longer appeared in the documentation on October 21. At these rates ML4 costs about 2.7 times as much as Mistral Large 3 ($0.50 input, $1.50 output). Because this is a preview, behaviour and pricing may change before general availability.
The licence for the open weights has not been published yet. Until the weights ship, Mistral's AI Act documentation lists only proprietary terms, and Mistral's documentation notes that on-premise deployments require a commercial licence, so check the terms once the weights are released. AppFor.ai has not tested the model.
Go to the source
Sources & further reading
Based on the official documentation below. Features and access can change; check the provider before making a decision.
- Mistral AI: Introducing Mistral Large 4 (opens in a new tab)
- Mistral Docs: Mistral Large 4 model page (opens in a new tab)
- Mistral Docs: changelog (October 6, 2026 entry) (opens in a new tab)
- Mistral AI: Mistral Large 4 technical documentation for downstream providers (AI Act) (opens in a new tab)
- Mistral Docs: Mistral Large 3 model page (opens in a new tab)
- The Frontier: Mistral Large 4 open-weights claims (opens in a new tab)
Was this page useful?
Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.