Mistral AI has unveiled its newest flagship language model. The company calls it Mistral Large 4. Internally, engineers nicknamed it “Le Chonk.” It entered public preview on October 6, 2026.
The model is available now through the Mistral Studio API. Its official model ID is mistral-large-4. Mistral positions it as Europe’s answer to frontier AI labs abroad.

Inside the Mistral Large 4 Mixture-of-Experts Design
Mistral Large 4 uses a granular Mixture-of-Experts architecture. The model card lists roughly 1.05 trillion total parameters. That scale puts it among the largest public models announced this year.
Active parameters per token differ slightly across Mistral’s own sources. The launch blog post cites 49 billion active parameters.
The published model card instead lists 52 billion. Both figures come directly from Mistral, so we note them rather than pick one as definitive.
The model is natively multimodal. A 1.6-billion-parameter vision encoder lets it read images alongside text. Output remains text-only for now.

Context Window and Training Infrastructure
Mistral’s model card advertises a one-million-token context window. That would let the model process entire codebases or lengthy legal filings.
However, one independent benchmarking site, Artificial Analysis, lists the current preview API at 524,288 tokens instead. That gap likely reflects preview-stage limits rather than a hard architectural ceiling.
Training ran on 3,800 Nvidia Grace Blackwell GPUs. All compute stayed inside Mistral’s own European data centers, according to Beam.ai. The model supports more than 160 languages, including every official European Union language.

Pricing, Open Weights, and Market Context
Preview API list pricing runs $1.36 per million input tokens. Output tokens cost $4.18 per million at list price. Multiple sources report a launch-sale discount near half that rate, roughly $0.68 input and $2.09 output, according to eesel.ai.
Open model weights are planned by the end of October 2026. Mistral has not announced an exact release date yet. The license for those weights also remains unpublished.
The launch follows Mistral’s recent valuation jump toward €21 billion. It also arrives amid a broader shift in how AI labs price their models.
Rivals like DeepSeek have pushed similarly large Mixture-of-Experts models at aggressive prices. That competitive pressure is reshaping the entire market.
Mistral’s open-weight tradition also matters here. The company’s commitment plays into ongoing debates over whether open-source AI can keep pace with closed rivals. A 1-trillion-parameter open release would be a serious test case.
Watch for Mistral’s official weight release before November. Check whether the final license permits commercial use freely. Compare actual preview-API context limits against the one-million-token claim once broader access opens. Those three signals will show whether “Le Chonk” lives up to its scale.
