← AI Hashrate 中文

MiMo-V2.5

Xiaomi · 310.0B (active 15.0B) · MoE · ctx 1024K · MIT · HuggingFace

MiMo-V2.5 (Xiaomi) is a 310.0B parameters MoE (15.0B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 174.5 GB VRAM for Q4 (weights 170.5 GB + KV 3.0 GB + 1 GB overhead) and 624.0 GB for FP16. About 3 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 339.4 tok/s (estimated); MI300X 192GB ≈ 224.8 tok/s (estimated); Mac Studio M2 Ultra 192GB ≈ 33.9 tok/s (estimated). Use the table below for VRAM fit and tok/s per dollar (list MSRP). Estimates follow memory-bandwidth math; measured rows override when present. See methodology for details. Methodology.

17 hardware configs — sorted by tok/s. Default fit context 8K.

HardwareVRAMQuanttok/sFits?tok/s/$
H200 141GB SXM5141Q4379.9 est.No0.011
B200 192GB SXM192Q4339.4 est.Yes0.008
MI300X 192GB192Q4224.8 est.Yes0.015
Gaudi 3 128GB128Q4213.7 est.No0.017
H100 80GB SXM580Q485.3 est.No0.003
A100 80GB SXM480Q451.9 est.No0.003
MacBook Pro M4 Max 128GB128Q435.6 est.No0.008
Mac Studio M2 Ultra 192GB192Q433.9 est.Yes0.005
Mac Studio M3 Ultra 96GB96Q430.0 est.No0.008
MacBook Pro M3 Max 128GB128Q426.1 est.No0.006
B200 192GB SXM192FP1625.2 est.No0.001
MI300X 192GB192FP1616.7 est.No0.001
RTX 6000 Ada 48GB48Q48.8 est.No0.001
L40S 48GB48Q47.9 est.No0.001
RTX A6000 48GB48Q47.0 est.No0.002
Mac mini M4 Pro 64GB64Q44.5 est.No0.002
Mac Studio M2 Ultra 192GB192FP162.5 est.No0.0

Fit checks

12 popular retail GPUs; every row in the table above also links its Fits? verdict to the full fit check.

Same family: MiMo

All MiMo models →