MiMo-V2.5 and V2.5-Pro
Xiaomi opened a 1.02T-total (42B active) Pro model and a 310B/15B model under MIT, with 1M context.
V2.5-Pro uses a 6:1 sliding-window-to-global attention ratio, three MTP layers, 27T-token FP8 pretraining, then agentic RL and multi-teacher on-policy distillation. API priced at $0.14/$0.28 (V2.5) and $0.435/$0.87 (Pro) per million tokens on OpenRouter.
- Date
- Wednesday, 22 April 2026
- Lab
- Xiaomi (MiMo)
- Kind
- open-weights
- Access
- open weights
- Price
- $0.435 input / $0.87 output per M tokens (Pro), OpenRouter
Figures
| Measure | Value | Measured by |
|---|---|---|
| Total / active parameters (Pro) | 1.02T / 42B | company |
| Context length | 1M tokens | company |
| KV-cache reduction from hybrid attention | nearly 7x | company |
API listing 2026-04-22 (OpenRouter, Wikipedia); open weights on HF from 2026-04-27 UTC. Benchmarks are Xiaomi-run.
Sources
- huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro
- openrouter.ai/xiaomi/mimo-v2.5-pro
- en.wikipedia.org/wiki/Xiaomi_MiMo
This record was checked against its sources on 6 October 2026. How we check