Mistral models in Big-AGI

Running Mistral in Big-AGI

Add your Mistral API key and use Mistral's models at Mistral's own API rates. Big-AGI isn't a reseller: no markup, no intermediary, Mistral bills you directly.

• Your key, your billing. Usage is billed by Mistral directly, nothing routes through us.
• Europe's frontier lab. EU headquarters, open weights for the majority of the lineup, and in-region European inference on Mistral's own platform.
• One clean model list. Mistral's API returns many aliases for the same model; Big-AGI folds them into one clean entry instead of a cluttered near-identical list.
• Hosted or local. Use Mistral's hosted API, or run its open-weight models locally through Ollama or LM Studio, and switch between them mid-chat.

Why not just Le Chat?

Le Chat is Mistral's model in Mistral's interface. Big-AGI runs Mistral next to GPT, Claude, Gemini, and anything else you've connected, side by side in the same conversation, which Le Chat isn't built to do.

It also opens up what Le Chat hides: temperature and top-p are real API parameters, not settings we invented. Your keys stay in your browser instead of a Mistral account.

The billing model changes too: Mistral charges per token, so a light user can spend less in a month than a subscription costs, and a heavy user never hits an artificial cap, just a bill for tokens used. Beam parallel runs do cost more, the honest tradeoff for comparing models side by side.

Your keys and your data

Turn on Direct Connection under Mistral's Advanced settings and the browser calls Mistral directly, bypassing the Big-AGI server, whenever your key is client-side and Mistral's API allows it. Your keys stay in your browser. Chats are stored locally first and sync only if you turn it on. The AI Inspector shows the exact request before it goes out: model, parameters, token counts, and a cost estimate.

Mistral in Beam

Beam runs Mistral against GPT, Claude, Gemini, or any other model you've connected, in parallel, then hands you Fusions: several strategies that combine, cross-check, and synthesize the parallel answers, which beats just picking the best one. Parallel runs use more tokens than a single chat.