Mistral AI has released Mistral Large 4, and the detail that sets it apart is not the size but the licence: the weights are coming.
What it is
Large 4 is a natively multimodal mixture-of-experts model with 1 trillion parameters, 49 billion of them active on any given token. It is available now as a public preview through the API, with open weights due at the end of October.
Pricing is $1.36 per million input tokens and $4.18 per million output. For comparison, OpenAI's GPT-6 Sol lists at $2 and $10, and Anthropic's Claude Sonnet 5.5 at the same $2 and $10.
What Mistral claims
The announcement leads with security: Mistral says Large 4 scores 82% on vulnerability reproduction and patching, which it calls the highest of any model. It also reports 61.7% on DeepSWE v1.1 for coding, and claims it outperforms all open-source models on HarveyAI's legal agent benchmark and exceeds GPT-6 Astra on legal and financial tasks.
These are the vendor's own numbers. Independent evaluations will follow once the weights are public, and that is when they are worth taking seriously.
Why open weights at this size matter
A model you can download is a model you can run inside your own infrastructure, fine-tune, and keep running if the vendor changes its prices or its terms. For regulated industries in Europe in particular, where data residency is a condition rather than a preference, a frontier-class model with published weights from a European company changes the procurement conversation.
The catch is hardware. A trillion parameters, even with 49 billion active, is not something you run on a workstation. Open weights at this scale mostly benefit organisations with their own GPU capacity, or cloud providers who will host it.
Who should look at it
Teams that need a strong model under their own control, and teams in legal, finance or security work, where Mistral is concentrating its claims. For everyone else, the API preview at $1.36 and $4.18 is simply one of the cheapest ways to try a frontier-sized model this month.