Re:MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per s...
by seewhydee on Monday June 08, 2026 @08:39PM
(#48454541)
That wouldn't explain why Deepseek is fast relative to other Chinese providers, especially considering that they're reportedly ahead of the curve among Chinese companies in moving off Nvidia. I think their quant fund background has more to do with it. Their models are clearly designed with performant inference clearly in mind.
Re:MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per s...
by ljosifov on Tuesday June 09, 2026 @01:56AM
(#48457071)
Yes, it's performant, and esp performant at non-trivial context depths. DeepSeek-V4 DS4 (and Flash - DS4F) drop tok/s speed much less than the rest. On my M2 Max it took context depths of 768K to drop tok/s to ~10 tok/s.
https://x.com/ljupc0/status/2062457314414587996
Other local models I've checked drop to unusable speeds way sooner. Only other model with similarity favourable curve I've tried is nemotron-cascade-2-30b-a3b. But it's a small model, way dumber than DS4F.
Coding agents use cases have large context depths. The rate of decline is as important as the headline number.
GamersNexus had a really good investigative piece (~3hrs long) on this where they went to China and met with grey market sellers. That piece absolutely pissed off NVidia and resulted in a fight with Bloomberg too.
Deepseek may be also be running inference on oodles of Chinese hardware but it wouldn’t surprise me for a second if they just acquired Blackwell chips through the grey market. The original Deepseek models were all trained using NVidia chips if I remember right.