MiMo-V2.6-Pro

MiMo-V2.6-Pro is Xiaomi's trillion parameter multimodal reasoning model with image, video, audio and text understanding, a 1M context window, tool calling and web search.

About MiMo-V2.6-Pro

MiMo-V2.6-Pro is Xiaomi MiMo's flagship reasoning model and part of the MiMo-V2.6 series released on September 22, 2026. Xiaomi describes the series as native fully multimodal models and documents MiMo-V2.6-Pro for complex projects, long horizon tasks, software engineering, cybersecurity, research and agent workflows.

The model supports joint understanding across text, images, video and audio. Its documented capabilities include multimodal perception, deep thinking, tool calling, streaming, web search, structured output and context caching. Xiaomi also demonstrates the V2.6 series performing computer use tasks by interpreting graphical interfaces and taking actions based on visual feedback.

MiMo-V2.6-Pro provides a 1 million token context window and supports up to 128,000 output tokens. The model is described as trillion parameter and was trained with large scale multi-task reinforcement learning covering code, general reasoning, visual tasks and cybersecurity.

The model is available through the Xiaomi MiMo Open Platform and Xiaomi MiMo API. Xiaomi has also released the model weights and technical report, allowing researchers and developers to access the open source model in addition to hosted API access.

Key Features

Native Multimodal Understanding: Supports joint understanding of images, video, audio and text.
1M Context Window: Supports up to 1 million tokens of context for long documents, multimodal inputs and extended workflows.
128K Maximum Output: Supports generation of up to 128,000 output tokens.
Deep Thinking: Includes reasoning capabilities for complex and multi-step tasks.
Agent and Tool Capabilities: Supports tool calling, web search and computer use scenarios involving graphical interfaces.
Developer Features: Supports streaming, structured output and context caching through the MiMo platform.

Pricing

Real-time API Input, Cache Hit: $0.0036 per 1 million tokens Applies to cached input tokens for MiMo-V2.6-Pro.
Real-time API Input, Cache Miss: $0.435 per 1 million tokens Applies to uncached input tokens for MiMo-V2.6-Pro.
Real-time API Output: $0.87 per 1 million tokens Applies to generated output tokens.
Batch API Input, Cache Hit: $0.0018 per 1 million tokens Batch processing rate for cached input tokens.
Batch API Input, Cache Miss: $0.2175 per 1 million tokens Batch processing rate for uncached input tokens.
Batch API Output: $0.435 per 1 million tokens Batch processing rate for generated output tokens.

Use Cases

Multimodal analysis across text, images, video and audio
Software engineering and long horizon coding tasks
Computer use and graphical interface automation
Research, reasoning and multi-agent workflows

Pros & Cons

Pros:

Supports native multimodal understanding across images, video, audio and text.
Provides a 1 million token context window with up to 128,000 output tokens.
Supports deep thinking, tool calling, web search, streaming, structured output and context caching.

Cons:

The hosted model is limited to 100 requests per minute and 10 million tokens per minute.
The documented output modality is text.

Integrations

Xiaomi MiMo API, Xiaomi MiMo Open Platform, Xiaomi MiMo Desktop

FAQ

What is MiMo-V2.6-Pro?

MiMo-V2.6-Pro is Xiaomi MiMo's flagship trillion parameter reasoning model in the MiMo-V2.6 series. It is designed for complex projects, long horizon tasks, multimodal understanding, coding, cybersecurity and research.

Is MiMo-V2.6-Pro a multimodal AI model?

Yes. Xiaomi describes the MiMo-V2.6 series as native fully multimodal and documents joint understanding of images, video, audio and text.

What is the context window of MiMo-V2.6-Pro?

MiMo-V2.6-Pro supports a context window of up to 1 million tokens.

What is the maximum output length of MiMo-V2.6-Pro?

MiMo-V2.6-Pro supports a maximum output of 128,000 tokens.

Does MiMo-V2.6-Pro support reasoning?

Yes. Deep Thinking is a documented capability of MiMo-V2.6-Pro, and Xiaomi describes it as a flagship reasoning model.

Does MiMo-V2.6-Pro support tool use?

Yes. The model supports tool calling and web search, and Xiaomi also demonstrates computer use capabilities involving graphical interfaces and visual feedback.

Does MiMo-V2.6-Pro support structured outputs?

Yes. Structured Output is listed as a supported capability for MiMo-V2.6-Pro.

Can MiMo-V2.6-Pro be used for coding?

Yes. Xiaomi's MiMo-V2.6 training includes code tasks, and the model is documented for software engineering and long horizon agent tasks.

Is MiMo-V2.6-Pro open source?

Yes. Xiaomi states that it has open sourced the MiMo-V2.6-Pro model weights and technical report along with reinforcement learning research resources.