Post #4381349
2026-08-04 18:09 UTC
I am just trying out DeepSeek v4 Flash 0731 and holly shit this thing is incredible for how small this model is and how little resources it needs.
It only needs 128 GB... which is at least 4 to 15 times less then the comparable new Models like GLM-5.2 or Kimi-K3...
I guess China will pop the American AI bubble sooner or later.... lets hope sooner than later.
#llm #ai #DeepSeek #v4flash #China
Replies (2)
-
@MaZderMind@chaos.social 2026-08-04 19:11
@m@lgbtqia.space We‘re running a deepseek v4 Flash and a QWen Instruct as for Vision Capabilities as a sidekick model. Together they are used for everything automation related and also some days f our employees (including me) use it as main AI model.
-
@m@lgbtqia.space 2026-08-04 18:18
The only good that these big American AI corpos provide is making it possible to distil their models with a lot of technical workarounds and to allow to get some very specific synthetic training data. Other than this these corpos have outlived their usefulness a while ago.