Back to AIBriefs
LaunchDevelopers

Fable 5 pushes Gemma 4 to 255 tok/s on WebGPU, releases demo

Fable 5 achieved 255 tokens per second inference speed for Gemma 4 on WebGPU using custom kernels. The project released the demo and kernels for local browser execution after claiming Anthropic's removal of safeguards enabled the optimization.

··Discuss
15 hours ago
Fable 5 pushes Gemma 4 to 255 tok/s on WebGPU, releases demo — AIBriefs