| Model profile | ||
|---|---|---|
| Model | Size | Approximate size |
| MiMo V2.6 ProModel source: provider1,048,576 tokens max native context | 1020Bactive 42B | 612 GB4-bit quantization |
| Compatible hardware Estimated speeds, not benchmark results: calculated from memory bandwidth and model size. Real results can differ significantly because there is no precise formula for deriving LLM generation speed from hardware specifications alone. | ||||
|---|---|---|---|---|
| Hardware | Memory | Bandwidth | vLLM | Estimated generation |
| DGX H200 | 1128 GBGPU HBM3e | 4800 GB/s | yes | 71 tok/s4-bit quantization |
| DGX Station | 748 GBCoherent Memory | 7100 GB/s | yes | 100 tok/s4-bit quantization |
| ET900N G3 | 748 GBCoherent Memory | 7100 GB/s | yes | 100 tok/s4-bit quantization |
| DGX B200 | 1440 GBGPU HBM3e | 8000 GB/s | yes | 111 tok/s4-bit quantization |
| GB200 NVL72 | 13400 GBGPU HBM3e | 8000 GB/s | yes | 111 tok/s4-bit quantization |
| GB300 NVL72 | 20000 GBGPU HBM3e | 8000 GB/s | yes | 111 tok/s4-bit quantization |
Teardown of Xiaomi MiMo V2.6 Pro and Flash configs reveals MoE routing, native 4-bit training, MTP layers, DFlash, GQA and KV cache mechanics enabling 1M context within 527 GB.
Tests MiMo V2.6 Pro on Mac Studio M3 Ultra using Inferencer v2.4.2, evaluating coding, agents, and multimodal tasks alongside Artificial Analysis Intelligence Index results.
Tests Xiaomi Mimo V2.6 Pro and Flash on website generation, C++ development, 3D modeling, Chrono City generation, RuneSClone, FPS development, and skate games.
Tests Xiaomi MiMo-V2.6 Pro across coding, frontend design, 3D game development, multimodal tasks, and computer use, comparing performance against other frontier AI models.
Tests MiMo-V2.6-Pro-RL, an open AI model designed to scale reinforcement learning toward self-improvement.