Apache 2.0 models on request
Mistral API:
which models you can use commercially.
Mistral AI publishes some models under Apache 2.0 and others under research or non-production licences. Before you pick one, check which group it is in. We can talk through adding the Apache 2.0 models to the Xinference Model API.
Licence by model
Mistral models, grouped by what the licence allows
| Model | Size | Context | Licence |
|---|---|---|---|
| Mistral Nemo Instruct | 12B | 128K tokens | Apache 2.0 |
| Mixtral 8x22B Instruct | About 141B (Mixture-of-Experts) | 65,536 tokens | Apache 2.0 |
| Mixtral 8x7B Instruct | About 47B (Mixture-of-Experts) | 32,768 tokens | Apache 2.0 |
| Mistral 7B Instruct v0.3 | 7B | 32,768 tokens | Apache 2.0 |
| Mistral Large 2 (2407) | 123B | 131,072 tokens | Mistral AI Research License: commercial use needs a separate licence from Mistral AI |
| Codestral 22B v0.1 | 22B | 32,768 tokens | Mistral AI Non-Production License: not for production use without a separate licence from Mistral AI |
Sources: Mistral AI’s model cards and licence tags on Hugging Face, and the Xinference model catalog for context lengths, checked 23 October 2026. This is a summary, not legal advice. Mistral, Mistral AI and the Mistral logo are trademarks of Mistral AI, shown only to identify the model. Their use does not imply endorsement.
Choosing among the Apache 2.0 models
Match the model to the workload
Long inputs: Mistral Nemo
A 12B model trained jointly by Mistral AI and NVIDIA, with a 128K context window. Small enough to run cheaply, long enough for big documents.
Harder tasks: Mixtral 8x22B
The largest Apache 2.0 Mistral model here. Mixture-of-Experts, so only part of it runs for each token.
Simple, high-volume calls: Mistral 7B
The smallest option for classification, extraction and routing where speed and cost matter most.
Mistral API questions
Which Mistral models are free for commercial use?
Of the models on this page, Mistral Nemo, Mixtral 8x22B, Mixtral 8x7B and Mistral 7B v0.3 are under Apache 2.0, which permits commercial use. Mistral Large 2 is under the Mistral AI Research License and Codestral 22B under the Mistral AI Non-Production License, so commercial use of those needs a licence from Mistral AI.
Can you add Mistral Large or Codestral?
Their licences restrict commercial and production use, so offering them depends on a commercial licence from Mistral AI. The Apache 2.0 models are the straightforward choice. Talk to us if you need something specific.
Is a Mistral API available on Xinference?
The Apache 2.0 Mistral models are available on request. The Model API’s standard models, with public per-token pricing, are Kimi K3 and DeepSeek V4.1 Flash.
Do you train on my data?
No. We don’t use your prompts or outputs to train or fine-tune models unless you agree in writing.
Last updated