Mistral rebuilds itself as a full AI stack around Large 4, with Vibe, Studio, Forge and Compute

Mistral is no longer pitching itself as just a model lab. Its new lineup spans the stack:

- Vibe, an agent for work and coding.

- Studio, for building applications.

- Forge, for customization and post-training.

- Compute, Mistral's own infrastructure offering.

All of it sits around Large 4, a 1.05-trillion-parameter model that has been in public preview since October 6. Mistral has promised open weights by the end of October, which means within days.

The expansion is backed by fresh capital. Mistral closed a €3B Series D at a valuation above €21B earlier this month, led by Samsung, Scaleup Europe and PSG Equity. It also shipped the Voxtral 3 speech family, with sub-500ms realtime transcription across 35+ languages. The company is extending into document AI and industrial engineering, including robotics and physics modeling. Its deepened NVIDIA partnership around European sovereign clouds, visible again at GTC Berlin, supplies the hardware side.

The strategic logic is that European sovereign-AI buyers want one accountable vendor, not a model plus three American clouds. Becoming full-stack lets Mistral capture compute margin rather than only API fees.

The risk is focus. OpenAI, Google and Microsoft each spend tens of billions per quarter on infrastructure, and Mistral is attempting the same breadth with a fraction of that. The near-term test is concrete: do Large 4's open weights actually ship by October 31 as promised?