reflex: a small open decision model with a WebGPU build

State plus typed questions in, calibrated probabilities out, built on Qwen3.5-4B, with a 650 MB WebGPU browser build on Qwen3.5-0.8B.

Notes

~100 ms after caching, 72% on MMLU for the 4B model, ECE 0.039 after tuning.

source-backed — Public repository, docs or live artifact. About this label

Destinations

Filed under

• #replica
• #webgpu
• #github

In collections

• Local & replica models →