The IT Home page is stamped 2026/10/21 10:55:37 and rests on Xiaomi's MiMo documentation. The judgment first: the offline time is written clearly, and no comparison score is printed on this page. The docs say the new versions have fully surpassed the old ones in effect and performance. Which tests, and by how much, this page does not say.

[1]
A closed binder
A closed binder, charcoal cover, muted blue spine, a small oxide-red edge on the pages., AI-generated illustration, not a news photograph

mimo-v2.5-pro and mimo-v2.5 went officially offline at 10:00 this morning. The table lists the offline time as 2026.10.21 10:00 (UTC+8). The system replacement time is 2026.10.14 18:00 (UTC+8). From 18:00 on 14 October 2026 (UTC+8), the two kinds of request route automatically: mimo-v2.5-pro to mimo-v2.6-pro, and mimo-v2.5 to mimo-v2.6-flash.

[1]

The two replacements are not the same. mimo-v2.5-pro accepted text only. After the switch to mimo-v2.6-pro it will support multimodal understanding, and the page says the API parameters are fully compatible. After mimo-v2.5 switches to mimo-v2.6-flash, default parameter values change.

[1]

In thinking mode, mimo-v2.6-flash and mimo-v2.6-pro do not accept custom temperature or top_p. The values actually passed are temperature 1.0 and top_p 0.95. If a mimo-v2.5 request had set its own parameters, those values carry over when the request routes to mimo-v2.6-flash. If max_completion_tokens is omitted, the system uses the flash default. On the table that default changes from 32768 to 131072. These are the defaults listed in the docs, not calling instructions.

[1]

Prices follow the figures from the 22 September launch. That day Xiaomi released and open-sourced the Xiaomi MiMo-V2.6 series. The page says it includes Pro and Flash, two natively multimodal models. Flash is 1 yuan per million tokens in and 2 yuan per million out. Pro is 3 yuan and 6 yuan, with a 99 percent cache discount. The page also says that, at a comparable level of capability, the price is 1/20 to 1/60 of overseas models. Both the comparable level and that range are claims from then. This page has no new comparison table.

[1]

A related-reading title says the series surpasses Kimi K3 and GLM-5.3 and ranks as the highest open model on the current AA index. The title prints no index score, and neither does the body. A title cannot be turned into a score.

[1]