Mistral AI

Mistral AI is a French AI company notable for both a hosted API (La Plateforme) and for releasing genuinely open-weight models under permissive licenses (Apache 2.0 for several lines) — downloadable and self-hostable, not just API-accessible, which distinguishes it from most other frontier-model vendors.

Two kinds of models: open-weight and proprietary

• Open-weight — e.g. the Mistral/Mixtral lines historically released with downloadable weights (including sparse mixture-of-experts (MoE) architectures like Mixtral, which route each token through a subset of expert sub-networks rather than the whole model — cheaper inference at similar quality to a much larger dense model). Runnable locally via Ollama/vLLM/llama.cpp or via any hosting provider, not just Mistral's own API.
• Proprietary, API-only — larger flagship models (historically branded "Large") offered only through La Plateforme or partner clouds, not released as downloadable weights.

Exact current model names, sizes, and licenses per release change — check Mistral's model list rather than assuming a specific line is still the current flagship or still open-weight.

Core API shape (La Plateforme)

from mistralai import Mistral

client = Mistral(api_key="...")
resp = client.chat.complete(
    model="mistral-large-latest",  # verify current model id against docs
    messages=[{"role": "user", "content": "Say hi in one word."}],
)
print(resp.choices[0].message.content)

The API deliberately mirrors the OpenAI Chat Completions shape (messages, role/content, tools for function calling, streaming via SSE) — most OpenAI-compatible client tooling works against it with only the base URL and model name changed.

Distinctive capabilities

• Function calling — same general JSON-schema tool-declaration pattern as other providers.
• Embeddings — a dedicated embeddings endpoint/model for retrieval/RAG use cases, separate from the chat models.
• Code-specialized models — a dedicated code-generation-focused model line alongside the general chat lines.
• Self-hosting — because several lines are open-weight, an organization with data-residency or sovereignty requirements can run a Mistral model entirely on its own infrastructure rather than calling a hosted API at all — a meaningfully different deployment option than most proprietary-only providers offer.

Common pitfalls

• Assuming every Mistral model is open-weight — only some lines are; others are proprietary and API-only. Check the specific model's license before planning to self-host it.
• Copying an OpenAI SDK snippet verbatim and expecting it to work unmodified — the request/response shape is deliberately similar but the client library, base URL, and some parameter names still differ.
• Hard-coding a -latest model alias in something that needs reproducible output — -latest aliases intentionally move to newer underlying models over time; pin a dated model id when reproducibility matters more than automatically getting improvements.

Learn more

• Mistral AI docs
• La Plateforme — API console and key management.
• Mistral model list