Xiaomi has permanently slashed prices on its in-house large language model API, the MiMo-V2.5 series, with cuts reaching as high as 99%. The overhaul also eliminates the premium for long-context inputs, unifying billing across all context lengths.
https://twitter.com/XiaomiMiMo/status/2059314052892099070
The flagship MiMo-V2.5-Pro model now charges just $0.0036 per million tokens when cache hits, a 99% drop from long-context pricing. Cache miss pricing is $0.435 per million tokens (down 78%), and output is $0.87 per million tokens (down 86%). For the standard MiMo-V2.5, cache hit pricing is $0.0028 per million tokens (down 98%), cache miss $0.14 per million tokens (down 83%), and output $0.28 per million tokens (down 93%).
Beyond per-token price cuts, Xiaomi revamped its subscription plans with a unified points system. The four monthly tiers remain at $6, $16, $50, and $100, but the available points have been boosted to 4.1 billion, 11 billion, 38 billion, and 82 billion, respectively. MiMo-V2.5 calls consume very few points — for example, a cache hit on MiMo-V2.5 uses just 2 points per token, a cache miss 100, and output 200. With the new efficient models, subscribers get up to 5–8 times more usable token capacity at the same price. All existing subscriptions have their points fully reset.
Alongside the price cuts, Xiaomi ended its earlier developer subsidy program ahead of schedule. The MiMo 100 Trillion Token Developer Initiative, launched on April 28, closed after approving 241,879 of 548,664 applicants, distributing 100 trillion tokens worth roughly 65.8 million yuan ($9.1 million). However, a long-term benefit for Apache Software Foundation committers continues unaffected.