Xiaomi MiMo API gets a permanent price cut, up to 99% off

Xiaomi MiMo announced a permanent price cut for its V2.5 series API, with up to 99% off and no more context window limits. The change mainly targets cached input scenarios, thanks to optimizations in the reasoning framework.

MiMo boss Luo Fuli explained the price drop comes from a layered KV cache optimization, boosting cached token capacity by 5 times and cutting cache costs by 80%. The model's architecture also keeps raw inference costs well below industry average, leaving room for the reduction.

With the new prices, MiMo's inference engine runs near full capacity but still breaks even. Luo believes affordable, high-performing model APIs will drive massive inference demand, boosting the whole AI infrastructure chain—from chips to data centers—and speeding up global AGI progress.