# MiMo V2.5 Pro

MiMo V2.5 Pro is Xiaomi's Pro-tier MiMo v2.5 model. It's aimed at workloads where multi-step reasoning, complex agent trajectories, and code accuracy matter more than peak throughput.

- **Model ID:** `xiaomi/mimo-v2.5-pro`
- **Type:** chat
- **Providers:** gmicloud
- **Context window:** 1,050,000
- **Maximum output tokens:** 131,000
- **Pricing:** $0.3045/1M input tokens, $0.609/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/mimo-v2.5-pro

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'xiaomi/mimo-v2.5-pro',
  prompt: 'Why is the sky blue?'
})
```

## About

MiMo V2.5 Pro is the Pro variant in Xiaomi's MiMo v2.5 family, released April 22, 2026 under the MIT license. Compared to the standard tier, Pro activates a larger share of a larger parameter pool per token, which raises reasoning depth at higher per-token cost.

Like the rest of the line, MiMo V2.5 Pro uses a Mixture-of-Experts (MoE) stack with hybrid attention. Sliding-window and full attention combine in a fixed ratio, which cuts KV-cache storage versus dense attention at the same sequence length. Three multi-token prediction (MTP) blocks raise output tokens per inference step. The full window of 1.1M tokens fits long agent traces, repos, or document sets.

MiMo V2.5 Pro supports reasoning, tool calling, file input, vision, and implicit prompt caching. Call it through GMICloud via AI Gateway. For lower-cost everyday work, see `mimo-v2.5`.

## What to consider

MiMo V2.5 Pro sits at the Pro end of MiMo v2.5. Per-token cost is higher than `mimo-v2.5`, but accuracy on hard math, agentic, and engineering tasks is the reason to pick it. Use AI Gateway's routing and fallback to send easy work to `mimo-v2.5` and reserve MiMo V2.5 Pro for the requests where reasoning depth pays off.

## When to use

### Best For

- **Long-horizon agents** with trajectories that span many tool calls in one session
- **Complex software engineering** including issue resolution, repo-level edits, and multi-file refactors
- **Math and proofs** that need long logical chains where intermediate steps matter
- **Long-context reasoning** over documents and codebases approaching 1.1M tokens
- **Pro-tier MoE compute** when you want the higher-capability tier in the MiMo v2.5 family

### Consider Alternatives When

- You're optimizing for cost and throughput. `mimo-v2.5` fits everyday agent and code work for less per token
- You only need short prompt-and-reply answers and don't benefit from deeper reasoning
- You want speed-first throughput. `mimo-v2-flash` from the previous generation is throughput-tuned
- The task is simple classification or extraction where a smaller model is cheaper and fast enough

## Best for

- **Long-Horizon Agents:** Trajectories that span thousands of tool calls in a single run
- **Complex Software Engineering:** Issue resolution, repo-level edits, and multi-file refactors
- **Math and Proofs:** Long logical chains where intermediate reasoning steps matter
- **Long-Context Reasoning:** Documents and codebases approaching 1.1M tokens
- **Pro-Tier MoE:** Higher active-parameter compute for harder reasoning workloads

## Consider alternatives

- **Throughput-Sensitive Workloads:** `mimo-v2.5` runs cheaper per token for everyday agent or code work
- **Short Prompt-and-Reply Calls:** A smaller model is enough when you don't need deep reasoning
- **Speed-First Tasks:** `mimo-v2-flash` from the previous generation is throughput-tuned
- **Simple Extraction Jobs:** A lightweight model handles classification at lower cost

## Frequently asked questions

### How does MiMo V2.5 Pro differ from `mimo-v2.5`?

It's the Pro tier. MiMo V2.5 Pro activates a larger share of a larger parameter pool per token than `mimo-v2.5`, with higher per-token cost in return for stronger reasoning, code, and agentic scores.

### What architecture does MiMo V2.5 Pro use?

A Mixture-of-Experts (MoE) stack with hybrid attention. Each token activates a subset of expert blocks, and sliding-window plus full attention combine to keep KV-cache storage manageable across the 1.1M tokens window.

### What's the context window for MiMo V2.5 Pro?

1.1M tokens. Hybrid attention keeps long-context runs practical, and multi-token prediction raises output tokens per inference step.

### Does MiMo V2.5 Pro support tool calling and reasoning modes?

Yes. MiMo V2.5 Pro supports reasoning and tool calling, both exposed through AI Gateway. Use them through the AI SDK, the Chat Completions API, the Responses API, or any other supported format.

### How do I authenticate requests to MiMo V2.5 Pro through AI Gateway?

Add your API key in AI Gateway project settings. Use `xiaomi/mimo-v2.5-pro` in API calls. AI Gateway routes, retries, and fails over across `gmicloud`.

### What does MiMo V2.5 Pro cost?

See the pricing section on this page for today's rates. AI Gateway tracks each provider's pricing for MiMo V2.5 Pro, so the numbers shown stay current.

### Does MiMo V2.5 Pro support zero data retention?

Zero Data Retention is not currently available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

### Can I route between MiMo V2.5 Pro and `mimo-v2.5` automatically?

Yes. AI Gateway supports fallback and routing. Send hard reasoning, code, and agentic requests to MiMo V2.5 Pro and fall back to `mimo-v2.5` for simpler tasks to keep costs in check.

## Links

- [Model page](https://vercel.com/ai-gateway/models/mimo-v2.5-pro)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://console.gmicloud.ai/user-console/ie/model-hub/llm/3e23bf2a-2353-4ad5-9e9a-ad04014e329f)