Mistral Large 4 is in preview on Complete, with open weights on the way
New models on Complete
Mistral Large 4 is Mistral's new flagship model, released as a public preview on October 6. It reads documents, charts and images alongside text, and Mistral says the open weights ship by the end of October. That will let companies run it on their own servers or private cloud. It is available for enterprise teams on Complete.
Mistral Large 4 preview | |
|---|---|
Independent score | 38 |
Cost per task | $1.13 |
Speed | 116 tokens per second |
Time to start answering | 1.5 seconds |
List price per million tokens in / out* | $1.36 / $4.18 |
Context window | 524,000 tokens |
Input | Text, documents, charts and images |
* Providers price per million tokens, which are fragments of words.
Independent scores
Mistral Large 4 scores 38 on the Artificial Analysis Intelligence Index for $1.13 per task. The index is an independent rating built from 10 tests covering agent work, coding, general knowledge and science.
On today's numbers, other models score higher at a similar or lower cost. Gemini 3.8 Flash scores 41 for $1.24 and Ling 3.1 Flash scores 41 for $0.99. MiMo-V2.6-Pro, an open model from Xiaomi, scores 46 for $0.13. Mistral says its reinforcement learning run is still in progress and expects further gains before the final release.
Fast to start
It starts answering in 1.5 seconds on Mistral's API, against a median of 3.8 seconds for comparable models on Artificial Analysis.
It writes at 116 tokens per second, against a median of 75.
Mistral's own results
Mistral reports these scores for the preview:
61.7% on DeepSWE, a software engineering test.
59.9% on AutomationBench, a test of 657 business automation workflows.
93% of 40 challenges on Cybench, a security test, and 82% on a test of reproducing and patching software vulnerabilities.
Second of five models in a human rated coding evaluation run by Surge AI, behind Claude Opus 5.
These are Mistral's results, not independent ones.
Open weights and European hosting
The case for Mistral Large 4 rests on control over where it runs.
Mistral says the open weights ship by the end of October. It has not yet named the license.
Once released, the model can run on premise or in a private cloud.
Mistral plans a European deployment that it runs end to end, under European law.
Training data covered more than 160 languages, including every official EU language.
Pricing
List price is $1.36 per million input tokens and $4.18 per million output tokens, with a 90% discount on cached input.
Run it on your own workflows
Mistral Large 4 preview is available for enterprise teams on Complete. Book a demo to run it against your own documents and workflows.
