Xiaomi MiMo V2.6 Provider for Claude Code

MiMo-V2.6-Pro is Xiaomi’s flagship reasoning model, with a 1M token context window and support for text, image, video, and audio input. MiMo-V2.6-Flash is the cheaper tier for high-volume work. Both use the Anthropic-compatible endpoint at api.xiaomimimo.com/anthropic, so Claude Code connects directly.

Template status: current. claude-multi 0.13.2 ships mimo-v2.6-pro[1m] and mimo-v2.6-flash[1m] in both the pay-as-you-go and Token Plan templates. The Fable alias maps to Pro. The templates set CLAUDE_CODE_MAX_OUTPUT_TOKENS to 128000 and API_TIMEOUT_MS to 3000000. Existing instances pick up the change with claude-multi doctor fix.

Model specs

V2.6 Pro handles complex projects, long-horizon agent tasks, and multimodal analysis. V2.6 Flash handles high-frequency calls, large batch work, and subagent traffic. Both output text only.

Xiaomi retired V2.5-Pro and V2.5 on October 21, 2026 at 10:00 UTC+8. The old IDs return errors, so a config created before that date will fail until the model names are changed.

Upgrading from V2.5

If you created a MiMo instance before October 21, 2026, its settings still name the retired V2.5 IDs. Run:

claude-multi doctor check
claude-multi doctor fix

The sync replaces the model slots, adds mimo-v2.6-flash[1m] for subagents, and maps the Fable slot to Pro. Both mimo and mimo-token instances use the same model IDs; only the base URL and key differ. The [1m] suffix is a Claude Code context hint, not part of the vendor model name. It tells the client the real 1M window instead of the 200K default.

Setup

1. Create an account at platform.xiaomimimo.com and generate an API key
2. Run the setup command:

claude-multi add mimo

3. Paste your API key when prompted

The template writes the base URL, V2.6 Pro and Flash model slots, and CLAUDE_CODE_MAX_OUTPUT_TOKENS=128000.

Pricing details

Xiaomi bills per token with no monthly minimum on the standard API. Cache hits are cheap enough that long, repeated context barely moves the bill.

A Pro cache hit costs about 0.8% of the input rate. A Flash hit costs about 2%. The pricing pages list no separate cache-write fee, so the first pass bills at the input rate and reuse bills at the hit rate.

At these rates, V2.6 Flash is one of the cheapest 1M-context models in claude-multi. It undercuts DeepSeek Flash on output at any hour, and its input rate is slightly lower even off-peak. Pro costs about 3.1x Flash on input and output.

Rates come from Xiaomi’s model pages for V2.6-Pro and V2.6-Flash.

Token Plan

Xiaomi also sells a Token Plan subscription with a monthly credit pool. Use the mimo-token template for that account type:

claude-multi add mimo-token

Token Plan runs on regional endpoints. The template defaults to China. After setup, replace ANTHROPIC_BASE_URL with the endpoint shown in your subscription console:

The Token Plan key is separate from a pay-as-you-go key, and the quota covers mimo-v2.6-pro and mimo-v2.6-flash. Individual plans start at $6/month for Lite. The plan is limited to coding tools, not automated scripts or application backends, so check the terms before pointing a production service at it.

Where it fits

MiMo is the low-cost multimodal option. If your agent needs to read screenshots, watch a screen recording, or take audio input, V2.6 Pro handles it in the same 1M window as the text context. Flash does the same for less when the task is routine.

The 128K output ceiling matches what Claude Code allows for an unrecognized model ID. DeepSeek and MiniMax list larger model maximums, but the client caps output at 128,000 either way, so the practical ceiling is the same.

Use the standard API for bursty work and the Token Plan when your monthly usage is predictable. The model IDs are the same on both, so switching plans does not require a new config beyond the base URL and key.

Related providers

• DeepSeek - also pay-per-token, 1M context, 384K output
• Kimi - 1M context on K3.1 Pro, built for tool use
• MiniMax - same 1M context, 512K output