Xiaomi MiMo
Listed at https://mimo.xiaomi.com
Overall Rank #42 ⭐ Consider
⚠️ Chinese company but no mainland-China endpoint on the OpenRouter route; proxy or latency measurement required. Xiaomi own portal at mimo.xiaomi.com may serve China traffic, not verified from our test environment. | 🌍 International
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | 按每百万 token 计费：mimo-v2.6-flash $0.14、mimo-v2.6-pro $0.435、mimo-v2.6-pro-ultraspeed $4.35、mimo-v2.5 $0.40、mimo-v2.5-pro $0.48 | per million tokens |
| Output | 按每百万 token 计费：mimo-v2.6-flash $0.28、mimo-v2.6-pro $0.87、mimo-v2.6-pro-ultraspeed $8.70、mimo-v2.5 $2.00、mimo-v2.5-pro $1.80 | per million tokens |
💡 Free Credits: No free tier; pure pay-as-you-go
🤖 Supported Models (5)
mimo-v2.6-flash (309B MoE / 15B active, 1M ctx, $0.14/$0.28, cached $0.0028)mimo-v2.6-pro (1T+ MoE / ~15B active, 1M ctx, $0.435/$0.87, cached $0.0036)mimo-v2.6-pro-ultraspeed (1T+ MoE, 1M ctx, $4.35/$8.70, cached $0.036, 1000+ TPS)mimo-v2.5 (omnimodal agentic, 1M ctx, $0.40/$2.00, cached $0.08)mimo-v2.5-pro (1T+ MoE, 1M ctx, $0.48/$1.80, cached $0.096)
✨ Pros
- ✓All five models carry a 1M-token context window, the largest in the 2026 API market
- ✓MiMo-V2.6-Flash at $0.14/M input is the cheapest MoE model with a 1M context window
- ✓384-expert MoE architecture (8 active per token), the largest expert count in a production API model
- ✓Native omnimodal input (text+image+video+audio), not bolt-on adapters
- ✓Open weights (MXFP4 on Hugging Face), providing a self-hosting exit path
- ✓UltraSpeed edition achieves 1000+ TPS for latency-sensitive workloads
⚠️ Cons
- ×Platform is only two months old (launched 2026-09-21), pricing is introductory and may change
- ×No mainland-China-optimized endpoint on the OpenRouter route, China latency needs measurement
- ×No Batch API discount or dedicated deployment tier, no volume discount for high-capacity offline work
- ×V2.5 output pricing is 5x input ($2.00/M), discouraging long-generation workloads
- ×Provider list is only five vendors (GMICloud, DeepInfra, NovitaAI, Xiaomi, StreamLake), limited redundancy
🎯 Best For
Teams that need a 1M-token context on a budget; developers who want open-weight self-hosting optionality; agentic workloads (tool use, codebase navigation); and scenarios needing omnimodal input without paying for the Gemini multimodal pipeline