MiMo-V2.6-Flash

MiMo-V2.6-Flash is the lower-cost model in 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 309 billion total and 15 billion active parameters and, like the Pro model, a 1-million-token context window, up to 128,000 output tokens, and text, image, video and audio input with text output. Xiaomi positions it for high-frequency calls and large-scale work. The overseas price list gives $0.14 per million input tokens, $0.28 per million output tokens and $0.0028 per million cached input tokens, with a batch API at half those rates; it shares a price row with mimo-v2.5, which is deprecated at 10:00 Beijing time on 21 October 2026, so a team calling mimo-v2.5 by name has a month to move. The weights were published on the model host on 21 September as MiMo-V2.6-Flash-RL, and, as with the Pro model, the card says the model is also available on the API without saying the two are identical. Artificial Analysis had not published a page for it when read on 22 September, so no independent score is recorded. Xiaomi's own figures (67.9 on DeepSWE v1.1, 28.8 on Terminal-Bench 4.0, 80.8 on OSWorld-Verified) are its own evaluation and are not stored as scores. The same release included MiMo-V2.6-Distill-Qwen-9B, a nine-billion-parameter model distilled onto another vendor's open base, with weights and no API, which Xiaomi offers as a starting point for research rather than a product.