Mistral launches Mistral Large 4 in preview as a one trillion-parameter model

Mistral says Mistral Large 4 has one trillion parameters and is its most capable model so far. The company has launched the model in preview for developers and says it will make it fully available later this month. Its claims about performance are company statements, with the model positioned for work including coding, agentic workflows and multimodal understanding.

Mistral calls the model le Chonk and says it can compete with the strongest open-source models globally. It also claims that the system significantly outperforms any open-weight model developed in the US or Europe. Those descriptions set out Mistral’s assessment; the source material does not provide independent test results or a comparison methodology.

Preview access precedes wider release

The current release is a preview for developers, according to Mistral. The company has not yet made the model fully available, and the supplied material gives no specific day for that planned release. Mistral says wider availability is expected later this month, keeping the launch at a preview stage for now.

Mistral also plans to offer the model through Mistral API. The source does not specify access terms, pricing, usage limits or which developer groups can use the preview. For teams assessing regulated technology, those details will shape how they can evaluate the system, but they remain unreported in the supplied material.

Mistral describes performance across tasks

The company says Mistral Large 4 shows performance across coding, agentic workflows and multimodal understanding. It also says the model performs strongly on enterprise workloads in cybersecurity, finance and law. These are Mistral’s claims about the system, rather than externally verified findings in the source material.

Mistral further says that visual grounding is a domain where the model surpasses some frontier closed models. The supplied account does not name the models used for comparison or report scores, test conditions or benchmarks. That leaves readers without the information needed to assess how broad or repeatable the stated advantage may be.

Funding accompanies the model launch

Mistral raised around $3.4bn in a Series D round in September, according to the source account. The round valued the company at around $24bn. Those figures place the model announcement alongside a major funding event, although the source does not say how the proceeds will be allocated or link the financing directly to Mistral Large 4.

The company has also begun making third-party models available to its customers, the source reports. It does not identify those models or explain how they relate to Mistral Large 4. The account describes Mistral as a European competitor to US AI labs and cheaper open-source Chinese models, but gives no comparative pricing or performance data for those alternatives.

Release details remain unconfirmed

Mistral’s stated plan is to make Mistral Large 4 fully available later this month through Mistral API. The supplied material does not confirm the release date, commercial terms or the model’s eventual access conditions. It also provides no response from Mistral to outside scrutiny, beyond the company’s own performance statements. The next update to follow is whether wider availability arrives as planned and whether further evidence supports the company’s comparisons.