Xiaomi MiMo-V2.5 Agent Model Guide: Pricing, Models, Claude Code, OpenCode
Xiaomi MiMo-V2.5 and V2.5-Pro now have sharply lower pay-as-you-go API pricing, 1M context, MIT-licensed weights, OpenAI/Anthropic-compatible APIs, and direct support for Claude Code and OpenCode.
Xiaomi MiMo-V2.5 is now worth tracking for agent builders. The new family includes MiMo-V2.5-Pro, a 1.02T-parameter MoE model with 42B active parameters and 1M context, and MiMo-V2.5, a native full-modal model for text, image, video, and audio understanding.
Xiaomi says the MiMo-V2.5 weights are open-sourced under the MIT license, and the API platform supports both OpenAI-compatible and Anthropic-compatible protocols. That makes MiMo unusually relevant for coding agents.
Official sources: MiMo-V2.5 open-source announcement, MiMo pay-as-you-go pricing, MiMo price adjustment announcement, MiMo Claude Code setup, and MiMo OpenCode setup.
Update for May 29, 2026: Xiaomi cut MiMo-V2.5 series pay-as-you-go pricing effective May 27, 2026 at 00:00 Beijing time. MiMo-V2-Pro and MiMo-V2-Omni will auto-route to V2.5 pricing on June 1, 2026 and are scheduled to be fully deprecated on June 30, 2026.
Model Lineup
| Model | Role | Context | Max Output | Best For |
|---|---|---|---|---|
mimo-v2.5-pro | Flagship agent/coding model | 1M | 128K | Long-horizon coding, autonomous agents |
mimo-v2.5 | Full-modal agent model | 1M | 128K | Text, image, video, audio workflows |
mimo-v2-flash | Low-cost text model | 256K | 64K | High-throughput coding and simple tasks |
API Pricing
Overseas API prices per 1M tokens:
| Model | Cache-Miss Input | Cached Input | Output | Long Context Above 256K |
|---|---|---|---|---|
mimo-v2.5-pro | $0.435 | $0.0036 | $0.87 | No separate 256K-1M tier after the May 27 cut |
mimo-v2.5 | $0.14 | $0.0028 | $0.28 | No separate 256K-1M tier after the May 27 cut |
mimo-v2-flash | $0.10 | $0.01 | $0.30 | Not applicable; 256K context |
Domestic API prices per 1M tokens:
| Model | Cache-Miss Input | Cached Input | Output | Long Context Above 256K |
|---|---|---|---|---|
mimo-v2.5-pro | ¥3.00 | ¥0.025 | ¥6.00 | No separate 256K-1M tier after the May 27 cut |
mimo-v2.5 | ¥1.00 | ¥0.02 | ¥2.00 | No separate 256K-1M tier after the May 27 cut |
mimo-v2-flash | ¥0.70 | ¥0.07 | ¥2.10 | Not applicable; 256K context |
Cache writing is currently free for a limited time. Web search is billed separately at $5 per 1,000 overseas calls or ¥25 per 1,000 domestic calls. The Token Plan billing system was also adjusted in the same announcement, with Xiaomi saying usable volume increased to roughly 5-8x depending on plan and scenario. Treat plan quotas as control-panel data and recheck before purchase.
Base URLs
Pay-as-you-go API:
OpenAI compatible: https://api.xiaomimimo.com/v1
Anthropic compatible: https://api.xiaomimimo.com/anthropicToken Plan, China region:
OpenAI compatible: https://token-plan-cn.xiaomimimo.com/v1
Anthropic compatible: https://token-plan-cn.xiaomimimo.com/anthropicToken Plan also exposes Singapore and Europe endpoints from the subscription console.
Claude Code Setup
For Claude Code, use the Anthropic-compatible endpoint:
{
"env": {
"ANTHROPIC_BASE_URL": "https://api.xiaomimimo.com/anthropic",
"ANTHROPIC_AUTH_TOKEN": "your-mimo-api-key",
"ANTHROPIC_MODEL": "mimo-v2.5-pro",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "mimo-v2.5-pro",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "mimo-v2.5-pro",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "mimo-v2.5-pro"
}
}For Token Plan, replace the base URL with your subscription endpoint and use the tp-xxxxx key from the subscription page.
OpenCode Setup
For OpenCode, create or edit ~/.config/opencode/opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"mimo": {
"npm": "@ai-sdk/openai-compatible",
"name": "MiMo",
"options": {
"baseURL": "https://api.xiaomimimo.com/v1",
"apiKey": "your-mimo-api-key"
},
"models": {
"mimo-v2.5-pro": {
"name": "mimo-v2.5-pro",
"limit": { "context": 1048576, "output": 131072 },
"modalities": { "input": ["text"], "output": ["text"] }
}
}
}
}
}If you need image understanding, use mimo-v2.5 and add image to the input modalities.
When to Try MiMo
| Scenario | Recommendation |
|---|---|
| You want an open-weight agent model with 1M context | Test mimo-v2.5-pro |
| You need multimodal agent perception | Test mimo-v2.5 |
| You need a cheap coding fallback | Test mimo-v2-flash or route simple multimodal work to the newly cheaper mimo-v2.5 |
| You need mature global enterprise API operations | Compare against OpenAI, Anthropic, and Google first |
Related: