Xiaomi MiMo-V2.5-Pro UltraSpeed Mode Hits 1000+ Tokens/s for Trillion-Parameter Models

Xiaomi and TileRT launched the UltraSpeed mode for MiMo-V2.5-Pro, achieving over 1000 tokens per second for trillion-parameter models on common GPUs. The mode boosts inference speed through optimized system and model co-design.

The UltraSpeed API is now available at 3x the price of MiMo-V2.5-Pro but offers about 10x faster output. For reference, MiMo-V2.5-Pro costs 0.025 yuan per million tokens for cached input and 3 yuan for uncached, with output at 6 yuan.

Due to limited high-speed resources, access is by application from June 9 to June 23, 2026. Approved users get free Chat access with up to 10 daily sessions, each capped at 30 minutes.