On request, licence permitting
MiniMax API:
check the licence before the benchmark.
MiniMax-M3, M2.7 and M2.5 are strong models for agents and long documents, but each ships under different commercial terms. Here is what each one allows, and how we can talk through adding the right one to the Xinference Model API.
Four models, three sets of terms
What each MiniMax model lets you do commercially
| Model | Size | Commercial use |
|---|---|---|
| MiniMax-M3.1-Flash | About 428B total, about 23B active | MiniMax Community License, same terms as M3: show “Built with MiniMax M3.1-Flash” and send MiniMax a one-time notice; above US$20M yearly revenue it needs MiniMax’s written authorisation. |
| MiniMax-M3 | About 428B total, about 23B active | MiniMax Community License. Commercial use must show “Built with MiniMax M3” and send MiniMax a one-time notice; above US$20M yearly revenue it needs MiniMax’s written authorisation. |
| MiniMax-M2.7 | About 230B | Non-commercial licence. Any commercial use needs MiniMax’s prior written authorisation. |
| MiniMax-M2.5 | About 230B | MiniMax Model License, which grants a royalty-free licence to use, modify and distribute. Read the full terms for conditions. |
Sources: MiniMax’s licence files for MiniMax-M3, MiniMax-M2.7 and MiniMax-M2.5, checked 23 October 2026. This is a summary, not legal advice. MiniMax and the MiniMax logo are trademarks of MiniMax, shown only to identify the model. Their use does not imply endorsement.
The newest one
What MiniMax-M3 brings
Native multimodal
MiniMax describes M3 as a native multimodal model, so text and images go into the same model rather than through a separate vision add-on.
1M-token context
Listed at 1,048,576 tokens (model maker’s figure), long enough for large codebases or a full set of contract documents in one request.
Mixture-of-Experts
About 428B parameters in total, with about 23B active for each token. Only part of the model runs per token, which is why a model this large can still respond at a workable speed.
MiniMax API questions
Can I use MiniMax models commercially?
It depends on the model. MiniMax-M3 and M3.1-Flash allow commercial use with a “Built with MiniMax” notice for the specific model and a one-time notice to MiniMax, plus written authorisation above US$20M yearly revenue. MiniMax-M2.7 needs written authorisation for any commercial use. MiniMax-M2.5 comes with a royalty-free model licence.
Which MiniMax model is best for coding agents?
MiniMax built M2.5 for coding and agentic workflows, and describes M2.7 as able to build complex agent harnesses. M3 adds image input and a 1M-token context, which helps when an agent has to read a lot at once.
Is MiniMax on the Xinference Model API?
On request. The Model API’s standard models are Kimi K3 and DeepSeek V4.1 Flash. For MiniMax, tell us the model and your expected volume. Whether we can offer a given version also depends on MiniMax’s licence terms for that version.
What is the MiniMax-M3 context window?
MiniMax lists M3 with a 1,048,576-token context window, and M3.1-Flash with up to 1,000,000 tokens. M2.7 is listed at 204,800 tokens and M2.5 at 196,608. These are the model maker’s figures; the limit we would serve is agreed when a model is added.
Last updated