They say on their page it is 1.6 billion credit and mimo v2.5 pro takes 2 credit per token, mimo v2.5 takes 1 credit per token but here is how they get you, cached token is still billed the same credit per round trip, absolutely not suitable for coding cli then, because every single one of them by design would keep going back and forth with toolcalls, that's how they work, normally inference providers charge 1% for the pre existing cached context, but Xiaomi takes the full amount, I did 10 small tasks like not even that deep, small tasks and it is already at 12 or so million credit used, it used probably under a million context tasks were that mini, like saying hello, and mv this folder around, write some sql etc, like 10 total prompts same session, credit cost keeps snow balling, they don't mention nothing of this sort in the token plan docs or anything anywhere, for a big task it would be what 200 million token uncached, so 400million credit if you used mimo v2.5 pro, so with max 100$ plan you can use it for 4 tasks PER MONTH, honestly get anything over mimo token/coding plan, 40m token task(input+output) would be like 400million, cache hit rate is avg 90%
[link] [comments]
