What is Mistral?
Mistral AI is a Paris-based frontier lab shipping open-weight and commercial LLMs — Mistral Large, Small, Codestral, Ministral — via La Plateforme at api.mistral.ai.
Resources
Short guides, curated links, a glossary and FAQ for Mistral AI and the LLM API ecosystem.
Quick guides
La Plateforme vs open weights
You can either call Mistral's hosted API (fast, EU-hosted, usage-billed) or self-host open-weight releases (Mistral 7B, Mixtral, Codestral Mamba) under an Apache-2.0 or research licence.
OpenAI-compatible mode
Mistral's chat/completions endpoint accepts the same JSON schema as OpenAI, so most SDKs work by swapping the base URL and API key. See /openai-compat for the mapping.
Reading a status page
A green dot means the endpoint is reachable. It does not guarantee model quality, rate-limit headroom, or per-region latency — those live on internal telemetry we can't probe from a browser.
Why CORS limits browser checks
api.mistral.ai does not return permissive CORS headers, so browser checks confirm DNS/TCP/TLS handshake — not the actual JSON response. That's why we report reachability, not full health.
Curated links
Official status pages
Open weights
Communities
Glossary
- La Plateforme
- Mistral's hosted API surface at api.mistral.ai — chat completions, embeddings, moderation, agents, fine-tuning.
- Mistral Large
- Mistral's flagship frontier model — best reasoning and multilingual performance in the family.
- Mistral Small
- Cheaper, lower-latency general-purpose model — good default for chat and RAG.
- Codestral
- Code-specialized model with fill-in-the-middle support; strong on 80+ programming languages.
- Ministral
- Edge-sized models (3B / 8B) designed for on-device and low-latency inference.
- Function calling
- Structured tool-use format where the model returns JSON matching a tool schema instead of free text.
- Embedding
- A fixed-length vector representation of text used for semantic search, RAG and clustering.
- Context window
- The maximum number of tokens a model can attend to in one request (input + output).
- Token
- The unit of text the model reads; roughly 3-4 characters of English per token.
- OpenAI-compatible
- A provider whose HTTP schema matches OpenAI's, so existing SDKs work by changing the base URL.
- Rate limit
- Per-key cap on requests-per-minute or tokens-per-minute; enforced by La Plateforme per tier.
- RAG
- Retrieval-Augmented Generation — inject retrieved documents into the prompt so the model can cite fresh context.
- Fine-tuning
- Continuing training on your own data to specialize a base model for a task.
- Open weights
- Model weights released under a licence that lets you download and self-host, e.g. Apache-2.0.
- MoE
- Mixture of Experts — architecture used by Mixtral where only a subset of expert sub-networks fire per token.
FAQ
Is Mistral down right now?
Check the live dashboard on the home page — it polls api.mistral.ai from your browser every 15 seconds. For regional incident detail, see status.mistral.ai.
How do I get an API key?
Sign up at console.mistral.ai, add a payment method, and create a key under API Keys. The free tier includes limited daily requests.
Which model should I use?
Start with Mistral Small for general chat and RAG, Codestral for code, and Mistral Large for hard reasoning or multilingual work.
Is Mistral OpenAI-compatible?
Yes — the /v1/chat/completions endpoint accepts the OpenAI schema. Point your SDK at https://api.mistral.ai/v1 and swap the API key.