French AI company Mistral announced the public preview of Mistral Large 4 on October 6. It is a major European entry in a frontier-model race that often attracts attention mainly around American and Chinese labs. The preview is available through Mistral's API; the company says the weights will be released by the end of October.
Model specifications and benchmarks
Mistral's launch announcement describes a mixture-of-experts model with approximately one trillion total parameters and 49 billion active parameters. Its model card lists 1.05 trillion total parameters, text and image input, and a 1 million-token context window. Active parameters are the subset used for a particular token, rather than the full model size.
On the Artificial Analysis Intelligence Index, the preview scores 38: level with GPT-6 Luna at maximum reasoning effort and slightly below DeepSeek V4.1 Flash at 39. Those figures describe that evaluation, rather than a guarantee of equal performance on your own coding or document tasks.
API pricing and the weight-release timetable
| Per 1 million tokens | Standard price | Launch price |
|---|---|---|
| Input | $1.36 | $0.68 |
| Cached input | $0.14 | $0.07 |
| Output | $4.18 | $2.09 |
The official changelog specifies a 50% discount for the first two weeks. Check the current pricing before budgeting a long-running workload. The planned open-weight release is separate from today's hosted preview, so local deployment must wait for the actual weight files and license.
For another recent frontier-model announcement, see our Gemini 4 Argon overview.
Adapted from the original Chinese article, published on October 7, 2026.
Comments NOTHING