Mistral Large 4 entered public preview on October 6 with an unusually specific promise attached: open weights by the end of October 2026. That deadline is now days away, making this the nearest expected release in the Frontrumors Model Guide.

What the weights drop would mean

A weights release would make Large 4 the strongest European open-weight option and give self-hosters a current-generation alternative to DeepSeek V4.1-Flash (MIT, September 10) and Meta's Muse Glimmer 30B (August). Preview API pricing sits around $1.36/$4.18 per million tokens, but the license terms attached to the weights matter more than the API price.

End of October was Mistral's own wording at preview launch. The only rumor here is the exact day.

Our advice from the Model Guide stands: evaluate the preview freely, but hold production commitments until the weights and final terms are public. If Mistral ships this week, Large 4 moves from Rolling Out to a fresh evaluation. If it slips, the preview starts to look like a placeholder.

Update, October 27

As of this morning there is still no release: Mistral's news page lists no weights post, and the newest repositories in the mistralai organization on Hugging Face are the Timestral time-series models, not Large 4. Mistral's launch post describes Large 4 as a natively multimodal hybrid instruct-and-reasoning MoE with about 1 trillion parameters, 52 billion of them active, and repeats the promise in plain words: 'We will release the weights by the end of the month.' It does not name a license. Artificial Analysis lists the preview at 38 on its Intelligence Index.

Update, November 1

The October deadline has elapsed. Mistral's current model card now names the Mistral Ecosystem Open Weight License and lists 1M context at $0.70/$2.10 per million input/output tokens. The card still says Public Preview. We have not independently verified the downloadable weights artifact, so this update neither confirms a completed weights release nor asserts that none exists. Earlier prices and benchmark figures above are historical snapshots.