Free Mistral API in 2026: Studio Free Mode, Codestral and Vibe

Mistral has two free doors and one paid one. The free doors are the Free plan of its consumer app (now called Vibe, formerly Le Chat) and a rate-limited free mode in Mistral Studio, its API console. The paid door is the pay-per-token API that most production use ends up on. Third-party hosts add a few more free routes to Mistral models. Vendor statements below were read on Mistral's own pages on 2026-11-05; limits for hosts come from the catalog rows linked beside them. Where Mistral does not publish a number, we say so instead of guessing.

The free API: Mistral Studio #

The catalog row for Mistral's Free plan records that the Free plan includes Mistral Studio "in Free mode with rate-limited API access and no card", and that monthly API credits are not included. Mistral does not print the Free-mode limits on a public page: the catalog notes that they are shown only inside the console, under Admin Panel, API, Limits. So the honest answer to "how many requests do I get" is: open that page after you sign up and read your own numbers. The pricing page says only that API models are priced per million tokens, with batch processing at 50% off, and it does not describe an API free tier, API rate limits or data-use terms for it.

Practical consequences:

• Plan for a low ceiling. Free mode is for trying models and building a prototype, not for serving users.
• No credits to burn. Without monthly credits, going over the Free-mode limits means upgrading to paid usage, not waiting for a refill.
• Check data terms yourself. We could not confirm from Mistral's pricing page whether prompts sent on the free API are used for training: its "Model training" row reads "Opt-out" without labelling the plans. Read the terms in the console before you send real data.

The free app: Vibe (Le Chat) #

Mistral's pricing page lists these features for the Free plan: "Limited messages and web searches", "Limited coding sessions", basic image generation and editing, web and mobile access and 100+ connectors, plus Mistral Studio access per its FAQ. No numbers are given for the limits, and Pro is described only as having "higher ceilings compared to Free". The app is a chat product: it is not an API key, and chatting there does not give you API access.

Codestral #

Codestral (version 25.08) is Mistral's coding model. Mistral's models page puts it in the "Premier" group with no licence stated, which means you cannot download and run it yourself. The pricing page we read does not mention free access to it, so we do not claim a free Codestral endpoint. What does exist: the Free-plan catalog row lists Codestral among the Studio models, so it is available inside the rate-limited free mode, and LLM7 serves `codestral-latest` on its free token tier (32K context, 60 requests a minute, 100,000 tokens a day shared across its models). For editor use, a free coding agent may fit better; see free AI coding agents.

Which Mistral models you can run yourself #

The models page lists these licences, which matter if you want to skip the API entirely:

Apache 2.0 allows commercial use, so Mistral Small 4 and the Ministral 3 models are the clean choice if you need to ship a product on a model you run yourself. Mistral Large 4 is open weight but under Mistral's own licence: read it before you use it commercially. We did not find a commercial-use statement for the free API on the pages we read, so for hosted use check the terms in your Studio account. To run the small models on your own machine for free, see run AI locally for free.

Mistral models through other free hosts #

This guide does not list an OpenRouter route for Mistral because we have not checked one today; the OpenRouter free models guide has the current list of `:free` models.

Which one to pick #

• You want Mistral's newest models via the official API: Mistral Studio in Free mode. Read your limits in the console and expect them to be modest.
• You need a Mistral model with no account on a vendor you do not know: Cloudflare Workers AI, which only needs a Cloudflare account.
• You want code completion from Codestral: the Studio Free mode, or LLM7's free token.
• You will ship a product: run Mistral Small 4 or Ministral 3 yourself under Apache 2.0, or move to the paid API.

Steps #

1. Create an account at the Mistral console (Studio) and choose the Free mode.
2. Open Admin Panel, API, Limits to read your limits.
3. Create an API key and point an OpenAI-compatible or Mistral SDK at it, with a model name from the console's model list.
4. Add retry with backoff on HTTP 429, since free-mode limits are tight.

For the other routes, create the account at the provider, copy its key, and use the model ID from its own page; the code is the same.

What we could not confirm #

A guide that hides gaps is worse than none, so here is what stayed open on 2026-11-05. Mistral's help-center article on the free API plan returned a 404 and the documentation page for usage tiers returned a 404 too, so we could not read a vendor statement on phone verification, free-mode request numbers or training on free-API data. The pricing page gave no numeric limits for the Free app plan. The models page lists version labels, not API identifiers for current models, so take the exact model name from your console's model list rather than from a blog post (including this one). When one of these gaps closes we will update the guide and its date.

How the three Mistral routes differ #

If you only want to compare answers between models, the app is enough. If your code has to call a model, you need a Studio key. If the code will be used by other people, plan for paid usage or a self-hosted Apache 2.0 model from the start, so you do not have to rewrite when the free limits run out.

Limits and the catch #

• No published numbers. Mistral hides the Free-mode limits behind the console, so any number you see in a blog post may be out of date.
• No monthly credits. The free API is a rate limit, not a balance.
• App and API are separate. The Vibe Free plan does not extend to the API beyond Studio's Free mode.
• Premier models cannot be self-hosted. Codestral and Mistral Embed are API-only.
• More guides: free AI API without a credit card, free embeddings API, free LLM APIs compared.