MiMo-V2.6-Pro

MiMo-V2.6-Pro is the flagship of Xiaomi's MiMo-V2.6 series, announced on 22 September 2026 with open weights under the MIT licence. It is a sparse mixture-of-experts model with 1.02 trillion total and 42 billion active parameters, a 1-million-token context window and up to 128,000 output tokens; it takes text, images, video and audio as input and returns text. Xiaomi's overseas price list puts it at $0.435 per million input tokens, $0.87 per million output tokens and $0.0036 per million cached input tokens, with cache writes free for a limited time and a batch API at half those rates. That is the same price, in the same row of the same table, as mimo-v2.5-pro — and the date that matters most for existing users is on that page too: mimo-v2.5-pro and mimo-v2.5 are deprecated at 10:00 Beijing time on 21 October 2026, and Xiaomi recommends switching as soon as possible. The price list also carries a separately billed "UltraSpeed" version of the same model at ten times the token rates ($4.35 in, $8.70 out), which Xiaomi describes as up to 20 times faster; that claim has not been measured here. The weights appeared on the model host at 15:39 UTC on 21 September under the name MiMo-V2.6-Pro-RL, and the card does not say whether the open checkpoint is identical to the model the API serves. Artificial Analysis, measuring it independently, scores it 46 on version 4.3.2 of its Intelligence Index — first of the 114 open-weight models on that index when read on 22 September — at 124.5 output tokens per second. Xiaomi's own figures (71.9 on DeepSWE v1.1, 34.9 on Terminal-Bench 4.0, 82.0 on OSWorld-Verified) are its own evaluation and are not stored as scores.