Mistral AI

It positions itself as the European / sovereign-AI alternative to US labs, with an open-weight-first strategy — as of September 2026 nearly its entire generalist lineup (Large 3, Small 4, the Ministral 3 family, Medium 3.5) ships downloadable weights under Apache 2.0 or a Modified MIT licence. Alongside the API it sells Studio, Forge, Vibe (agent/coding), Mistral Compute, and on-prem/cloud deployments via Azure, AWS Bedrock, Vertex AI, Snowflake, IBM watsonx and Outscale.

A Access and limits

Sign up at console.mistral.ai / admin.mistral.ai to create a workspace and API key; the API is at https://api.mistral.ai/v1 and the chat-completions surface is OpenAI-SDK compatible (change base URL + model name). There is a free experimentation tier with restrictive rate limits; Mistral no longer publishes exact free-tier RPS/token numbers — current limits are shown per-workspace at admin.mistral.ai/plateforme/limits. Pricing modifiers documented officially: Batch (/v1/batch) = 50% off list; cached input tokens = 90% off input (exact per-model cached prices are published in the docs pricing table); Priority Tier = 1.75x list pricing (a 75% premium on input, output and cached tokens) with a 99.5% uptime SLA and custom, model-specific limits; Regional inference = 1.1x list pricing (10% surcharge).

Endpoints: global api.mistral.ai (no upcharge), EU api.eu.mistral.ai and US api.us.mistral.ai (both +10%); regional endpoints support function calling only — Agents, Batch and Files APIs are unavailable there. Enterprise APIs (regional data-processing controls, system-level SLAs, higher rate limits, premium support) are quoted at 75% above list. Agent/tool add-ons are billed separately on top of model tokens: code execution $30/1K calls, web search $30/1K calls, image generation $100/1K images, premium news $50/1K calls, data capture $0.04/M tokens, Libraries OCR $3/1K pages + indexing $1/M tokens + $0.01/call.

Fine-tuning is offered through the Classifier API: one-off training $1/M tokens with a $4 minimum per job, $2/month/model storage, then inference at $0.1/$0.1 per M tokens on the Ministral 3B classifier and $0.04/$0.04 on the 8B classifier. Note the whole Pixtral, Devstral, Magistral, Mixtral, Mistral Nemo, Mistral Saba and Codestral Mamba lines have been fully retired — the last of them (Mistral Medium 3 / 3.1, Devstral 2, Magistral Medium 1.2) reached their retirement dates on or before 31 Aug 2026, so nothing from those families is still served.

C Calling the API

Mistral AI accepts requests in the OpenAI Chat Completions format, so any OpenAI-compatible client works by changing the base URL. Example uses ministral-3b-2512.

curl https://api.mistral.ai/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ministral-3b-2512",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

from openai import OpenAI

client = OpenAI(
    base_url="https://api.mistral.ai/v1",
    api_key=os.environ["API_KEY"],
)

response = client.chat.completions.create(
    model="ministral-3b-2512",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

A compatibility layer is not always the provider's full API — caching, strict tool schemas and structured output are commonly unsupported there. Check the official docs before relying on an advanced feature.

D Official references

• Homepage
• API documentation
• Pricing