Gemma 4 WebGPU kernels reportedly reach 255 tokens per second
Gemma 4 reportedly reached 255 . If can run above 100 , local private models become more practical for everyday AI work.
Routine analysis, drafting, and repeated agent steps could run locally, while Claude or Codex would be saved for harder reasoning tasks. The main point is that a fast can reduce API token use and cost while keeping more data on the user’s own machine.
Key points
- Gemma 4 were reported at 255 .
- Over 100 on could make useful for more daily work.
- Local private models may reduce paid API calls for repeated .
- Claude or Codex could be reserved for harder reasoning instead of every step.