📢 Update: Mistral Medium 3.5 is now available — 128B dense model replacing Medium 3.1 and Devstral 2. See the Medium 3.5 complete guide, how to run it locally, and API guide.
Mistral AI is a Paris-based AI company founded in 2023 by former DeepMind and Meta researchers. In under three years, they’ve reached a $6 billion valuation, hit 1 million Le Chat downloads in two weeks, and built models that compete with GPT-5 and Claude at a fraction of the cost.
They’re the only major AI lab based in Europe, which matters for GDPR compliance and data sovereignty.
The model lineup
| Model | Parameters | Best for | Price (API) |
|---|---|---|---|
| Mistral Large 2 | 123B dense | Complex reasoning, coding | $2/$6 per 1M |
| Devstral 2 | 123B dense | Agentic coding (72.2% SWE-bench) | $2/$6 per 1M |
| Devstral Small 2 | 24B dense | Local coding agent | Free (Apache 2.0) |
| Codestral | 22B dense | Autocomplete, FIM | $0.30/$0.90 per 1M |
| Mistral Small | 22B dense | Fast, cheap general tasks | $0.10/$0.30 per 1M |
| Mistral Nemo | 12B dense | Edge deployment | Free (Apache 2.0) |
Why Mistral matters
For developers:
- Models are 3-7x cheaper than Claude/GPT at comparable quality
- Devstral 2 matches Claude Opus on SWE-bench (72.2%)
- Open-weight models you can self-host and modify
- Codestral is the best autocomplete model available
For European companies:
- GDPR-compliant infrastructure based in the EU
- Data stays in Europe — no transatlantic transfers
- The only frontier AI lab subject to EU regulation by default
- Partnerships with Capgemini, Stellantis, TotalEnergies
For the open-source community:
- Apache 2.0 license on key models (Devstral Small, Nemo)
- Weights available on Hugging Face
- Runs locally via Ollama
Products
Le Chat — Mistral’s AI assistant (like ChatGPT). Free tier + Pro at ~$15-25/month. Includes Vibe CLI for terminal coding.
La Plateforme — Developer API. Pay-per-token, starting at $0.10/1M tokens for Mistral Small.
Pricing vs competitors
| Task | Mistral | OpenAI | Anthropic |
|---|---|---|---|
| Complex coding | $2/1M (Devstral 2) | $10/1M (GPT-5.4) | $15/1M (Opus) |
| General chat | $0.10/1M (Small) | $2.50/1M (GPT-4o) | $3/1M (Sonnet) |
| Autocomplete | $0.30/1M (Codestral) | $0.15/1M (Mini) | N/A |
Mistral is consistently 3-7x cheaper for equivalent quality.
FAQ
Is Mistral AI open source?
Partially. Some models are fully open under Apache 2.0 (Devstral Small, Mistral Nemo), while others like Mistral Large 2 and Devstral 2 use more restrictive licenses that allow commercial use but with conditions. Mistral’s approach is to open-source smaller models while keeping their largest models under controlled licenses.
Why would I choose Mistral over OpenAI or Anthropic?
Three main reasons: cost (3-7x cheaper for equivalent quality), European data sovereignty (GDPR compliance without transatlantic data transfers), and open-weight models you can self-host. If you’re a European company with data residency requirements, Mistral is the only frontier AI provider based in the EU.
What’s the difference between Mistral’s coding models?
Codestral (22B) is for fast autocomplete and tab completions in your IDE. Devstral 2 (123B) is for complex agentic coding tasks like refactoring and feature implementation. Use Codestral as your inline suggestion engine and Devstral 2 as your coding agent — they’re designed to complement each other.
Related: Mistral Large 2 Complete Guide · Devstral 2 Complete Guide · Codestral Complete Guide · AI and GDPR for Developers