MiMo-V2.5

Overview

MiMo-V2.5 is an open-weight native omnimodal model from Xiaomi MiMo that entered public beta on April 23, 2026. It combines a 310-billion-parameter sparse Mixture-of-Experts (MoE) language backbone with dedicated vision and audio encoders to understand text, images, video, and audio in long-context agent workflows.

Key Features

• Native Omnimodal Understanding: Processes text, image, video, and audio input in one model, using a 729M-parameter vision encoder and a 261M-parameter audio encoder in addition to the language backbone.
• 310B Sparse MoE Architecture: Uses 310 billion total parameters, 15 billion active parameters, 256 routed experts, and eight selected experts per token.
• Efficient 1M-Token Context: Interleaves 39 Sliding Window Attention layers with nine global-attention layers in a 5:1 pattern. Xiaomi reports nearly 6x lower KV-cache storage while retaining long-context performance.
• Multimodal and Coding Evaluation: Xiaomi reports 77.9 on MMMU-Pro, 87.7 on Video-MME, 23.8 on Claw-Eval Multimodal, 65.8 on Terminal-Bench 2.0, and 56.1 on SWE-Bench Pro.
• Agent-Oriented API Controls: Supports deep thinking, streaming, function calling, structured output, and Xiaomi's separately billed web-search service.

Best Use Cases

• Multimodal Analysis: Understanding screenshots, charts, scanned documents, video, and audio alongside written instructions and contextual data.
• Visual and Media-Aware Agents: Agents that must inspect multimodal evidence, reason about it, call tools, and produce a text response or action plan.
• General Coding and Automation: Software tasks that benefit from image-based UI context, terminal tools, structured output, and lower API cost than the Pro variant.
• Long-Context Applications: Document analysis, codebase assistance, media archives, and extended conversations that need up to a one-million-token context.

Capabilities and Limitations

Capability	Description
Reasoning	Deep thinking is enabled by default and can be turned off with thinking.type. Xiaomi's official multimodal results include 77.9 on MMMU-Pro and 81.0 on CharXiv RQ.
Coding	Xiaomi reports 71.8 on MiMo Coding Bench, 62.3 on Claw-Eval Text, 65.8 on Terminal-Bench 2.0, and 56.1 on SWE-Bench Pro.
Creative Writing	Supports general and long-form text generation.
Multimodal	Accepts text, image, video, and audio input and produces text output. Official results include 87.7 on Video-MME, 83.5 on DailyOmni, and 23.8 on Claw-Eval Multimodal.
Context Window	1M tokens.
Max Output	128K tokens.
Tool Use	Supports function calling, structured output, streaming, and Xiaomi's web-search tool. In thinking-mode agent conversations, tool-call history must retain the complete reasoning_content field.
Multilingual	The official model repository identifies English and Chinese support.

Known Limitations

• The hosted API provides multimodal understanding with text output; it does not generate images, audio, or video as native response modalities.
• In thinking mode, custom temperature and top_p values are ignored; the API forces 1.0 and 0.95. Multi-turn tool workflows that omit historical reasoning_content can fail with HTTP 400 or lose context quality.

Pricing

Model	Input (Credits/Token)	Cache Write (Credits/Token)	Cache Read (Credits/Token)	Output (Credits/Token)	Web Search (Credits/Use)
MiMo-V2.5	0.14	0.14	0.0028	0.28	-

Pricing note

Prices shown in the documentation are B.AI standard reference prices for base billing purposes. B.AI may provide lower actual usage costs through top-up bonuses and account benefits. Specific prices, bonus Credits, and account benefits are subject to the platform display and final billing records.