Benchmark

Every model, on this device, in this browser.

detecting… ← Studio

Runs each model 10× on the bus sample after a warm-up and reports the median inference time (model only, excluding pre/post-processing). Models download on first use and are cached afterwards.