作者： Jakub Rusinowski · 最后更新： 2026年9月6日
Xiaomi's MIT-licensed MoE pair, and the clearest licence story at the trillion-parameter tier — MIT, not a bespoke open-weight agreement with revenue thresholds. Both carry 1M-token context and agentic multimodal capability. Consistently omitted from best-of roundups despite matching Kimi K2.6 on aggregate index scores.
| Licence | What it permits | Applies to |
|---|---|---|
MIT | Commercial use permitted Commercial use permitted. No usage restrictions beyond attribution. | MiMo-V2.5 310B, MiMo-V2.5-Pro 1T |
| MiMo-V2.5 310B | Min 188 GB VRAM · Q4_K_M · 1,000,000 ctx · |
| MiMo-V2.5-Pro 1T | Min 605 GB VRAM · Q4_K_M · 1,000,000 ctx · |
The cheapest GPU that runs MiMo-V2.5 locally (min 188 GB VRAM) is the Apple M2 Ultra (192 GB).
Install Ollama then run: ollama run
Minimum VRAM: 188 GB. For best results use Q4_K_M quantization.
MiMo-V2.5 needs about 188 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: MiMo-V2.5 310B (188 GB, Q4_K_M); MiMo-V2.5-Pro 1T (605 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
MiMo-V2.5's smallest variant needs about 188 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
Q4_K_M is the best balance of quality and VRAM for MiMo-V2.5 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
Install Ollama, then run: ollama run . This downloads MiMo-V2.5 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.