What happened
On October 6, 2026, Mistral released Mistral Large 4 as an API preview. It is a multimodal mixture-of-experts model with about 1.05 trillion total parameters and 52 billion active. The company promised open weights by the end of October (the license was unspecified at launch; Hugging Face listed an expected date of October 31).
Pricing and reported numbers
- $1.36 per million input tokens and $4.18 output;
- self-reported scores: 61.7% on DeepSWE v1.1, 28.3% on Terminal-Bench 4, 82% on a vulnerability reproduce-and-patch test;
- according to Simon Willison, its Artificial Analysis score jumped from 9 (Large 3) to 38, still behind the strongest frontier models.
Mistral trained it on its own cluster of 3,800 Grace Blackwell GPUs.
Why it matters
European open models are back in the upper tier that Chinese models had been dominating. For data-sovereignty or on-premises needs, the weights date matters more than the API preview.
What to do now
- Try the API preview on code and security tasks before moving large workloads.
- Wait for the weights license before planning local deployment.
- Compare cost and quality with GPT-6.1 Sol and Haiku 5.5 on your workload.