Qwen3.5-2B 2B params · gated DeltaNet ×18 + gated attention ×6 · custom WGSL · · bench
tok/s
Qwen3.5-2B
the first hybrid gated-DeltaNet LLM in the browser · WebGPU · zero dependencies
downloading the model — 1.07 GB
one time only: next visits load from the browser cache