Qwen3.8-27B vs MiMo-V2.6-Pro

Pyyan / Compare / Qwen3.8-27B vs MiMo-V2.6-Pro

2 of 5

Open-Weight Models · verified 29 Sept 2026

Qwen3.8-27BAlibabacurrent

XMiMo-V2.6-ProXiaomicurrent

Add another3 slots left

Specification	Qwen3.8-27B	MiMo-V2.6-Pro
Summary	27B dense under Apache 2.0, and nearly five million downloads.	A trillion parameters under the MIT licence.
Context	262K native, ~1M extended	1M
Input ($/Mtok)	Weights only	Weights only
Output ($/Mtok)	Weights only	Weights only
Licence	Apache 2.0	MIT
Max output	32K	Not published
Parameters	27B dense	1.02T total, 42B active
Released	5 Aug 2026	21 Sep 2026
Model ID	Qwen/Qwen3.8-27B	XiaomiMiMo/MiMo-V2.6-Pro
Category	Open-Weight Models	Open-Weight Models
Official	—	Xiaomi ↗

Highlighted rows are where these differ.

27B dense rather than a mixture of experts, so it behaves predictably on one machine
Apache 2.0, weights only, no hosted price
One of the most downloaded models on Hugging Face, with quantised builds everywhere

Best for running locally.

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.