Is there a way to test this in a browser interface? I'm not going to run any executables from any of the frontier labs on my machine, be that Xiaomi or Google or Anthropic.

Leaning into what it cost to train is hilarious and an obvious shot at US frontier labs spending tens to hundreds of millions or more to train their models.

As for the stats that everyone wants:

MiMo-V2.6-Flash-310B-A15B roughly GPT-5.6 Luna / Claude 4.9 according to benchmarks MiMo-V2.6-Pro-1.02T-A42B roughly GPT-5.6 Sol / Opus 5 according to benchmarks

Perhaps with IQ2 flash will run on 128G M5?

it is pretty amazing to see a young & talented woman is the tech lead of this frontier model.

it is reasonable to question why America doesn't have such environment.

does anyone know what unnamed model is on paretto frontier picture right between MiMo 2.5 and 2.6?

so weird to acknowledge someone being on the front edge, but not name it

These benchmark are useless as they don't say whether they were done before or after Fable and Astra got nerfed.

Experiences so far with the non-flash model: a bit of an overthinker/overplanner. Makes bad/messy architecture choices with an open-ended prompt. But actually very competent at intricate code level things, and pretty good at finding corner cases, including in code that other more advanced models had looked at and not noticed problems with.

I deliberately left things wide open and ambiguous to see what it would do. I think with better upfront planning this would be excellent value.

[flagged]

[flagged]

[flagged]