Xiaomi MiMo V2.5 Pro
MiMo V2.5 Pro
A trillion-parameter model under an MIT licence, from a phone company. That is where 2026 got to.
Updated
The read
MiMo V2.5 Pro is a very large sparse model with a small active parameter count, published under MIT, natively multimodal and unusually efficient at agentic work. Xiaomi is not the lab anyone expected this from, and the model is better than that framing suggests. Serving it is a real undertaking, and the ecosystem around it is younger than the licence generosity deserves.
Where it shines
- MIT licence on a model of genuinely frontier scale
- Small active parameter count keeps inference affordable
- Native multimodal input rather than a bolted-on encoder
Worth knowing first
- Tooling and community support lag the better-known families
- Serving the full model needs substantial GPU memory
Index score
82.4
Provisional. This score is read from the product's public capability, not from the one-prompt rebuild the fully reviewed entries went through, so treat it as a placing rather than a verdict.
- Rank
- 16 of 53
- Builds
- LLMs
- Category
- Open weights
- Output
- MIT-licensed weights, plus a hosted API
- Pricing
- Free tier
- Best for
- Agentic work on weights you are free to modify
Scores are our own editorial judgement, weighted craft 55, speed 10, control 15 and value 20.
Also worth a look
Full index- 05Kimi K3Kimi K3ProvisionalThe largest open-weights model anyone has shipped, and it reads a million tokens without blinking.Open weightsFree tier87.5
- 06GLM-5.2GLM-5.2ProvisionalFrontier-class output, an MIT licence, and a million tokens of context. That combination is the story.Open weightsFree tier87.1
- 12Qwen 3.6Qwen 3.6ProvisionalThe open Qwen release most people should actually run, in sizes that fit real hardware.Open weightsFree tier83.3
