Quick answer
Mistral Large 4 entered public preview on October 6, 2026. It is a granular mixture-of-experts model of about 1 trillion total parameters with roughly 49 to 52 billion active (sources differ) and a 1.6B vision encoder. List price is $1.36 per million input tokens and $4.18 per million output, with a launch sale at $0.68 and $2.09 for an unstated period; cached input is $0.14. Mistral says open weights will be released by the end of October; at preview it is API-only. Context is 1M tokens per Mistral's spec sheet, about 524K per independent evaluators, with a 262K output limit. The independent Artificial Analysis index scores it 38, behind GLM-5.3 (45), Kimi K3 (44), and DeepSeek V4.1 Flash (39).
Mistral is the European answer to the Chinese open-weight giants, and Large 4 is its first trillion-parameter-class model. The important word in the announcement is "preview": the API is live, the weights are not.
What is confirmed
- API availability from October 6 through Mistral's platform
- Pricing: $1.36 / $4.18 list, half price during the launch sale
- Architecture: granular MoE, roughly 1T total, around 50B active, vision built in, tool calling
- Open weights promised "by the end of this month"; one outlet reports October 27
What to treat carefully
Mistral's benchmarks are self-reported. The one independent reading available, the Artificial Analysis index, puts it behind the current Chinese open models. The context window is quoted at 1M by Mistral and measured at roughly half that by evaluators, with a 262K output cap. And the launch discount has no end date, so budget at list price.
Where it fits
At list price it undercuts the $2/$10 tier held by Sonnet 5.5, GPT-6.1 Sol, and Gemini 4 Argon, while costing several times more than DeepSeek V4.1 Flash. Its case is jurisdiction and vendor: an EU-hosted provider with enterprise contracts, and weights you will be able to run yourself. If that is what you need, it is the obvious candidate; if it is not, DeepSeek is cheaper and open today.
Bottom line
A credible European flagship at a good price, with the weights still to come. Test it on the API now; decide once the weights and an independent benchmark or two are public.
