Xiaomi cuts MiMo-V2.5 API prices up to 99%
Xiaomi has permanently cut MiMo-V2.5 API prices by up to 99% and raised Token Plan usage five- to eightfold, crediting inference work that reduces KV cache traffic and serving cost per token.
Xiaomi announced a permanent price cut for its MiMo-V2.5 model series API, effective globally from 0:00 Beijing time on May 27, 2026. The company says prices fall by up to 99% against the old rates, and that pricing no longer changes with input length. The page carrying the announcement is titled as a release of three MiMo-V2.6 models, but the text gives no details about any V2.6 models.
The subscription Token Plan now gives five to eight times as much usage at the same price, and Xiaomi says the billing rules are simpler. All active Token Plan quotas were fully reset on May 27 under the new rules, whatever users had already consumed. Xiaomi says it will announce gifts for past paid users whose plans have expired. The page does not list the new per-token prices.
Xiaomi also ended its creator incentive program early. The program started on April 28 and gave out its full pool of 100 trillion tokens by 16:08 on May 26. A separate benefit for Apache Software Foundation members remains open.
Xiaomi credits the price cut to inference optimizations. Its team added full support for sliding window attention (SWA) on top of SGLang HiCache. Xiaomi says this cuts KV cache data transfer between GPU memory, CPU memory and SSD to roughly one-seventh of the previous level, and lets the system cache nearly five times as many tokens. Xiaomi also says it raised cluster input throughput by changing its expert-parallelism scheme and how it groups requests by input length.