Pyyan / Compare / Edge0-145B vs MiMo-V2.6-Pro

Edge0-145B vs MiMo-V2.6-Pro

2 of 5

Open-Weight Models · verified 13 Nov 2026

EEdge0-145BEdge0current
XMiMo-V2.6-ProXiaomicurrent
3 slots left
SpecificationEdge0-145BMiMo-V2.6-Pro
SummaryA 145B model with about 10B active per token that Edge0 says runs in 29GB on a laptop.A trillion parameters under the MIT licence.
Context256K1M
Input ($/Mtok)Weights onlyWeights only
Output ($/Mtok)Weights onlyWeights only
LicenceEdge0 Community Model License 1.0 (Qwen terms apply)MIT
Max outputNot publishedNot published
Parameters145B total, about 10B active1.02T total, 42B active
Released12 Nov 202621 Sep 2026
Model IDEdge0/Edge0-145B-q3kXiaomiMiMo/MiMo-V2.6-Pro
CategoryOpen-Weight ModelsOpen-Weight Models
OfficialEdge0 ↗Xiaomi ↗

Highlighted rows are where these differ.

Edge0-145B

  • Weights on Hugging Face in Q3_K (40.81 GiB) and NVFP4 (47.79 GiB) builds, with Edge0-120B alongside
  • Edge0 reports 29GB peak memory, over 80 tokens a second and 86.52 on Terminal-Bench 2.1. These are its own figures, and its technical report is not yet out
  • The Hugging Face build instructions target NVIDIA CUDA 13. Edge0's blog claims day-one macOS support, while the card says its Apple measurements do not establish macOS availability
  • Licence is Edge0 Community Model License 1.0 for Edge0's contributions, with Qwen Community License terms still applying to Qwen materials
  • Selling the model, running a commercial model service or shipping a paid coding or office assistant on it needs a separate written licence from Edge0

Best for trying a large coding model locally on a 48GB laptop or a single 24GB GPU with host RAM.

Full spec sheet →

MiMo-V2.6-Pro

  • 1.02T total parameters, 42B active per token, a 4.1% activation ratio
  • Text, image, video and audio in, with a 1M token context
  • Shipped with 7,000+ reinforcement learning environments and the framework that trained it

Best for multimodal work on your own cluster.

Full spec sheet →