Skip to content
Learn · Guide · Mistral AI

Mistral

A European AI lab known for strong open-weight models you can self-host, plus Le Chat, a fast assistant with a real data-control story.

Mistral AI8 min readmistral.ai

What is Mistral?

Mistral AI is a Paris-based AI lab that builds both open-weight models you can download and run yourself and a hosted assistant called Le Chat. It's the most prominent Western source of high-quality open models.

The portfolio spans open-weight models (small enough to self-host) and larger commercial models available by API, plus Codestral for code. Because key models are downloadable, Mistral is a natural fit when you need to run AI on your own infrastructure for cost, latency, or data-residency reasons.

If you want frontier-ish quality without sending data to a US hyperscaler, or you want to own the model that runs your product, Mistral is the one to evaluate.

Strengths

What it's best for

  • Self-hosting: open-weight models you can run on your own servers for control over cost, latency, and data.
  • Data-sensitive deployments where keeping inference in your own environment (or a European region) matters.
  • Cost-efficient, fast inference: strong quality-per-dollar, especially with the smaller models.
  • Code generation with Codestral, a model specialized for programming.
  • Building products on an API from a vendor independent of the big US clouds.
  • Everyday assistant use through Le Chat (chat, web search, document upload, image generation).
Limits

Where it falls short

  • Absolute top-of-leaderboard reasoning. The largest closed models from OpenAI, Anthropic, and Google still tend to lead the hardest benchmarks.
  • The widest consumer feature set. Le Chat is leaner than ChatGPT or Gemini.
  • Teams with no appetite to manage models, if they choose the self-hosted path (the API avoids this).
How to use it

Two ways in: Le Chat or the weights

For a normal assistant, use Le Chat at chat.mistral.ai: sign up and chat, with web search, document upload, and image generation available.

For engineering, you have a choice unique among this group: call the models via Mistral's API (La Plateforme), or download the open-weight models and run them yourself on your own hardware or cloud.

How to use it

Self-hosting the open models

Open-weight Mistral models are published on hubs like Hugging Face and can run via common runtimes (for example vLLM or Ollama). This puts inference entirely inside your environment, and nothing leaves your network.

That control is the headline benefit: predictable cost at scale, low latency, and data that never touches a third party, for the same reasons teams self-host databases.

How to use it

Codestral and the API

Codestral targets code completion and generation and is designed to slot into developer tooling. The API also exposes the general and larger commercial models with per-token pricing.

Pick the smallest model that passes your evals. Mistral's smaller models are cheap and fast, and often enough for classification, extraction, and routing.

Pricing

What Mistral costs

Approximate, in USD, as of January 2026. Prices change often. Confirm on the official site before you rely on them.

Open weights

$0 (self-host)

Download and run the open models yourself; you pay only for your own compute.

Le Chat Free

$0

Hosted assistant with limits: chat, web search, and document upload.

Le Chat Pro

~$15 / month

Higher limits and access to the more capable models in the assistant.

API (La Plateforme)

Usage-based

Per-token pricing across the model range; small models are inexpensive.

Visit the official Mistral site
Try it

Example prompts

Copy these into Mistral as starting points, then adapt them to your task.

Pick a model for a job

I need to classify support tickets into 8 categories at high volume and low cost. Which Mistral model should I use, and write me a tight system prompt for it.

Code with Codestral

Write a Python function that validates and normalizes phone numbers to E.164, with tests for the tricky cases. Explain the edge cases you covered.

Extract structured data

From the text below, extract a JSON object with fields: company, role, location, salary_range. Return only valid JSON, null for anything missing.

Choose self-host or API

We process about 2 million short classification requests a month and care about data residency. Walk me through whether to self-host an open Mistral model or use the API, with the cost and operational trade-offs.

FAQ

Mistral
common questions.

Direct answers to the questions we get asked the most. If yours isn't covered, write to the team.

Work with SDEN

Putting AI into production?

We help teams choose the right models and ship them securely, self-hosted when data demands it. And we hand you the keys to run them in-house.

Mistral guide · SDEN