Xiaomi MiMo V2.5 API on TheRouter: Unified Routing for Pro, Flash & ASR Models (2026)

TheRouter now routes traffic to Xiaomi MiMo — Xiaomi's first-party large-model platform — as a vendor-direct provider. The MiMo family sits alongside the existing qwen/, glm/, and kimi/ Chinese first-party reasoning brands under TheRouter's standard catalog, with open-weights variants published on HuggingFace at XiaomiMiMo.

What ships today

Four new standard models under the xiaomi/ brand namespace:

• xiaomi/mimo-v2.5-pro — flagship reasoning model. Multimodal input (text + image + audio + video), text output. Strong math, code, and agent performance at a small-parameter footprint. Vendor-direct USD cost: $0.435 / $0.87 per million input / output tokens, with a ~99% cache-hit discount on input reads.
• xiaomi/mimo-v2.5 — mid-tier reasoning model balancing cost and capability. Same multimodal input surface as the Pro tier. Vendor-direct USD cost: $0.14 / $0.28 per million input / output tokens.
• xiaomi/mimo-v2-flash — fast, low-cost text model for high-volume workloads. Vendor-direct USD cost: $0.10 / $0.30 per million input / output tokens.
• xiaomi/mimo-v2.5-asr — speech-to-text transcription priced per audio duration. Vendor-direct USD cost: $0.074 per hour (≈ $0.001233/min ≈ $0.0000206/s) — significantly below comparable Whisper-tier routes for the same audio surface.

The ASR model plugs into the existing /v1/audio/transcriptions endpoint and is charged through TheRouter's duration-based audio ledger.

Pricing positioning

For Chinese first-party reasoning, the catalog now reads:

MiMo enters as the lowest-cost vendor-direct option in its tier.

Reasoning is off by default — opt in explicitly

MiMo exposes a non-standard thinking toggle that the gateway sets explicitly to disabled for every chat-completion request that doesn't opt in. Customers paying for text-output volume won't be surprised by accidental reasoning charges (reasoning output bills at the standard output rate per Xiaomi's published pricing).

Opting in is identical to every other reasoning-capable model TheRouter already routes. Use any one of:

• reasoning_effort: "low" | "medium" | "high" (OpenAI o-style)
• Anthropic-style thinking: { type: "enabled", budget_tokens: ... }
• Bailian / Qwen-style enable_thinking: true
• Zhipu thinking config

The adapter mirrors usage.completion_tokens_details.reasoning_tokens into the gateway-wide usage.reasoning_tokens field, so existing reasoning-token dashboards work unchanged.

Promo caveat — cache writes are free for now

Xiaomi's platform banner notes that cache writes are temporarily free as of 2026-06-06. TheRouter's cost catalog models cache-hit read pricing at the published per-million-token rate, and does NOT charge a cache-write surcharge. If Xiaomi ends the promo, the catalog will be updated and customers will see the new line item before any billing impact.

Routing details

xiaomi/* is a brand-new namespace; no other provider currently routes these models, so all four standard models land at priority: 0 (vendor-direct, sole route). Any future aggregator addition for the MiMo family will sit at strictly higher priority numbers — vendor-direct always wins by default.

How to use

Call https://api.therouter.ai/v1/chat/completions as you always would — set model to any of the four. No SDK change needed:

curl https://api.therouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "xiaomi/mimo-v2.5-pro",
    "messages": [{"role": "user", "content": "Explain in one sentence why the Riemann hypothesis is hard."}],
    "reasoning_effort": "high"
  }'

ASR goes through /v1/audio/transcriptions:

curl https://api.therouter.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -F model="xiaomi/mimo-v2.5-asr" \
  -F file=@meeting.mp3

All four xiaomi/* models honor TheRouter's existing provider.only / provider.ignore / provider.order controls if you need to pin or bypass the MiMo direct route.