Xiaomi open-sources MiMo-V2.6 omnimodal AI model family

Xiaomi said the Pro and Flash models are native omnimodal, accepting text, image, video and audio within one model family and supporting 1-million-token context windows. Pro's published architecture lists a 681-million-parameter vision encoder, a 308-million-parameter AudioTokenizer and a 127-million-parameter audio patch encoder capable of distinguishing speech.

The design targets agentic use cases and task instructions, particularly computer use. Xiaomi said a computer-use agent can read a task instruction, read a screenshot of a user interface or a video, reason about it and then decide what to do next inside a single model loop. A coding or design agent can combine source code and a lengthy repository context with screenshots, UI mockups and reference images in the same session. For business use, Xiaomi said the model can ingest numerous documents, including multimodal components such as call audio, screenshots, logs and notes, without separate tools to break down transcripts or orchestrate each modality. In its announcement, the company demonstrated the model on building game worlds, Blender-based 3D modeling, embodied simulation and video-music workflows.

According to SiliconANGLE's report, V2.6-Pro scored 46.32 on the Artificial Analysis Intelligence Index, ahead of Kimi K3 and Qwen3.8 Max and the highest-ranked open-source or open-weight model in that comparison at launch. It still trails the most advanced proprietary flagship models, Claude Fable 5.1 and GPT-6 Astra, on aggregate measures. On agentic tasks, V2.6-Pro came close to or occasionally passed leading closed systems: 53.1 on AutomationBench against Claude Opus 5's 50.3, 31.6 on Agents' Last Exam to tie Opus 5, and 89.9 on Terminal Bench 2.1 against Opus 5's 89.1.

Xiaomi said the V2.6 series retains the standard application programming interface pricing used for MiMo-V2.5, and the model cards list availability through the company's AI Studio, MiMo Desktop and MiMo Code. Pro and Flash are also offered on OpenRouter with 1.05-million-token context through the unified API. Flash is priced at $0.14 per 1 million input tokens and $0.28 per 1 million output tokens, Pro at $0.435 and $0.87, and Pro-UltraSpeed at $4.35 and $8.70 for the 20-times-faster generation.

Xiaomi claims Pro costs roughly one-twentieth to one-sixtieth as much as overseas models at comparable intelligence once cached-token costs, which the company says are significantly reduced, are taken into account. It cited $10 to $50 per 1 million input and output tokens for OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1. SiliconANGLE reported that the gap widens further for cache-heavy agents, while the relevant business comparison rests less on price per token than on price per task, which depends on reliability and how many loops were needed to finish the task successfully.