Xiaomi introduces Mimo-V2.6 series open-source AI model family
SiliconANGLE Kyt Dotson ● Covered by 4 sources
Xiaomi just open-sourced MiMo-V2.6, a new AI model family that takes text, images, video and audio. It says the bigger one can go toe-to-toe with top closed models on some agent tasks, while costing much less.
Based on reporting by SiliconANGLE, Kyt Dotson — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Xiaomi has pushed out MiMo-V2.6, an open-source generative AI family built around two natively omnimodal models: a flagship Pro version and a smaller Flash variant. The company is also shipping a slimmer Pro-UltraSpeed model that it says can generate output up to 20 times faster than Pro at the same quality.
The omnimodal bit is the real story here. Xiaomi says Pro and Flash can take text, images, video and audio in one model, with 1 million-token context windows. The published architecture for Pro includes a 681 million-parameter vision encoder, plus a 308 million-parameter AudioTokenizer and a 127 million-parameter audio patch encoder that can distinguish speech. That setup is aimed at agentic work: reading instructions, looking at screenshots or video, and deciding what to do next without bouncing between tools.
Xiaomi also pitched the model as useful for coding, design and business workflows. In its announcement, the company showed the model handling game worlds, Blender-based 3D work, embodied simulation and video-music tasks. It also said the system can ingest long, messy inputs like call audio, screenshots, logs and notes without needing a separate pipeline to split everything apart.
On benchmarks, Xiaomi said V2.6-Pro scored 46.32 on the Artificial Analysis Intelligence Index, ahead of Kimi K3 and Qwen3.8 Max. That still leaves it behind the leading proprietary systems on the market overall, but the agentic results are more interesting: 53.1 on AutomationBench versus Claude Opus 5’s 50.3, 31.6 on Agents’ Last Exam to Opus 5’s 31.6, and 89.9 on Terminal Bench 2.1, just above Opus 5’s 89.1.
The company says pricing will stay aligned with MiMo-V2.5, and the models are listed through AI Studio, MiMo Desktop and MiMo Code. Pro and Flash are also on OpenRouter through a unified API with 1.05-million-token context. Xiaomi prices Flash at $0.14 per 1 million input tokens and $0.28 per output, Pro at $0.435 and $0.87, and Pro-UltraSpeed at $4.35 and $8.70. It claims Pro lands at roughly one-20th to one-60th the cost of overseas models at comparable intelligence, depending on cached tokens and the task.
My take — AI-written commentary, not fact-checked reporting
Open-sourcing a model family is nice; open-sourcing a model family that can eat text, video, audio and screenshots in one go is the part that actually matters. Xiaomi is clearly aiming at the boring but valuable bit of AI: getting useful work done for less money, not winning a demo contest. That’s the pressure point for the big closed players, and it’s the one they hate most.
Read more about this at: SiliconANGLE
Related stories
Z.ai Releases GLM-5.3-Flash: A 320B-A18B Natively Multimodal MoE With a 1M-Token Context
MarkTechPost · 3 weeks ago ·
5
Moonshot AI Releases Kimi K3: A 2.8 Trillion Parameter Open MoE Model With Kimi Delta Attention and 1M Context
MarkTechPost · 2 months ago ·
9
Expanding Together AI Model Library into multimedia generation with 40+ new image and video models
Together AI · 11 months ago ·
52