MiMo V2.6 Flash: Xiaomi's Smaller Open-Weight Omnimodal Model

What MiMo V2.6 Flash is

MiMo V2.6 Flash is the smaller of the two models in Xiaomi's MiMo V2.6 series, announced by Xiaomi on September 22, 2026 alongside MiMo V2.6 Pro. Xiaomi describes both as natively omnimodal, and the Hugging Face model card lists text, image, video and audio as input types. Xiaomi's announcement says the Flash model "strikes the best balance between intelligence, efficiency, and cost" in the series; that is Xiaomi's description. The weights are published under the MIT licence.

Architecture

The model card describes MiMo V2.6 Flash as a sparse mixture-of-experts model with 309 billion total parameters and 15 billion active. It has 48 layers: 39 use sliding-window attention and 9 use global attention. The sliding-window layers have 64 query heads and 8 key-value heads; the global-attention layers have 64 query heads and 4 key-value heads. Each MoE layer has 256 routed experts with 8 activated per token and no shared experts. The context length is 1M tokens. The published checkpoint is the MOPD upgrade of the MiMo-V2.6-Flash-RL checkpoint.

Pricing

Xiaomi lists MiMo V2.6 Flash at $0.14 per million input tokens on a cache miss and $0.28 per million output tokens, with cache hits at $0.0028 per million. Cache writes are free for a limited time, according to the announcement, and Xiaomi states the V2.6 series keeps the API pricing of the V2.5 series. For comparison on the same price table, MiMo V2.6 Pro is listed at $0.435 input and $0.87 output per million tokens.

Training run

Xiaomi streamed the reinforcement-learning run live. Its announcement states that in under six days MiMo V2.6 Flash completed 30 RL steps over roughly 750,000 trajectories at a cost of about $0.85 million, against about $2.62 million for the Pro model. Xiaomi reports that the average pass rate on the training tasks rose by 25% in relative terms for Flash, and that on DeepSWE v1.1, a held-out long-horizon software engineering benchmark, the Flash model's score rose from 48.8 to 65.68 over the run. These are Xiaomi's figures.

What Xiaomi says the series can do

Xiaomi's announcement describes the V2.6 series combining 3D spatial reasoning, multimodal perception and computer-use agent capabilities. Its examples include building interactive 3D scenes from an image, video or text prompt, generating 3D objects in Blender, controlling a simulated Franka Panda robotic arm from multi-view camera feeds, producing frontend interfaces and slide decks, and composing music as MIDI. Some of the published case studies name the Pro model specifically; the announcement does not break the examples down by model in every case.

Availability

Xiaomi lists MiMo V2.6 Flash as available in AI Studio, MiMo Code and MiMo Desktop, through the MiMo API Platform, and on OpenRouter. The open weights are on Hugging Face under the XiaomiMiMo organisation, and the technical report, training environments and RL code are published alongside them.

Official Resources