Mistral Models
The complete Mistral AI model lineup: 18 models spanning the Mistral Large flagship, Mixtral mixture-of-experts series, Codestral coding models, and lightweight Devstral variants. Europe's leading AI lab, known for open weights and efficient architectures. Compare scores, API pricing, context windows, and capabilities. Updated hourly from live data.
All Mistral Models - Ranked by Score
18 models from Mistral AI, sorted by composite score
Mistral Pricing Comparison - Sorted by Output Price
| Model | Input $/1M | Output $/1M |
|---|---|---|
| Mistral Nemo | $0.019 | $0.030 |
| Mistral Small 3 | $0.050 | $0.080 |
| Ministral 3 3B 2512 | $0.100 | $0.100 |
| Ministral 3 8B 2512 | $0.150 | $0.150 |
| Ministral 3 14B 2512 | $0.200 | $0.200 |
| Mistral Small 3.2 24B | $0.094 | $0.250 |
| Voxtral Small 24B 2507 | $0.100 | $0.300 |
| Mistral Small 3.1 24B | $0.351 | $0.555 |
| Mistral Small 4 | $0.150 | $0.600 |
| Saba | $0.200 | $0.600 |
| Codestral 2508 | $0.300 | $0.900 |
| Mistral Large 3 2512 | $0.500 | $1.50 |
| Mistral Medium 3.1 | $0.400 | $2.00 |
| Mistral Medium 3 | $0.400 | $2.00 |
| Mistral Large 2407 | $2.00 | $6.00 |
| Mistral Large | $2.00 | $6.00 |
| Mixtral 8x22B Instruct | $2.00 | $6.00 |
| Mistral Medium 3.5 | $1.50 | $7.50 |
Understanding the Mistral Model Lineup
European AI Leadership
Mistral AI is Europe's leading AI company, headquartered in Paris. Founded by former DeepMind and Meta researchers, Mistral has rapidly become a major force in the AI landscape by releasing high-quality models with open weights. Their commitment to open-source and European AI sovereignty has made them a preferred choice for organizations seeking alternatives to US-based AI providers.
Mixtral MoE Architecture
Mistral pioneered the mainstream adoption of Mixture-of-Experts (MoE) architecture with Mixtral. MoE models use multiple specialized sub-networks ("experts") and a routing mechanism that activates only a subset for each token, delivering the quality of a much larger model at a fraction of the compute cost. This makes Mixtral models exceptionally efficient for both inference and fine-tuning.
Open Weights Philosophy
Mistral releases many of their models with open weights under permissive licenses, enabling self-hosting, fine-tuning, and community experimentation. Models like Mixtral 8x22B and Mistral 7B have become staples of the open-source AI ecosystem. Codestral and Devstral extend this philosophy to code-specialized models, providing strong coding assistants that can run on your own infrastructure.
API vs Self-Hosted
Mistral models are available both through Mistral's own API (La Plateforme) and major cloud providers including AWS Bedrock and Azure. For self-hosting, open-weight models can be deployed via vLLM, TGI, or Ollama. The prices shown here reflect current per-token API rates. Self-hosting eliminates per-token costs but requires GPU infrastructure.
相关页面
Explore Mistral comparisons, rankings, and pricing across the full model landscape.
Mistral offers a range from the efficient Mistral Small to the powerful Mistral Large. The lineup includes Mixtral (mixture-of-experts architecture), Codestral (coding-specialized), and Pixtral (multimodal).
Mistral offers both open-source and proprietary models. Mistral 7B and Mixtral 8x7B are open-source. Mistral Large and Codestral are available through their commercial API.
Mistral Large competes with GPT-4o on many benchmarks at lower prices. Mixtral offers strong performance for open-source deployments. Check our leaderboard for the latest benchmark comparisons.