On-device machine learning
Browser AI benchmark
Run real image-classification inference with MobileNet v2 in this browser. The benchmark compares the universal CPU/WebAssembly backend with WebGPU acceleration when the browser exposes it.
ONNX inference test
MobileNet v2 AI performance
Unlock the browser AI benchmark
Checking paid access...
Payment is processed securely by Stripe. One purchase unlocks both premium tests. By continuing, you agree to the Terms and Privacy Policy.
Ready to benchmark local browser inference.
The first run downloads a 13.3 MB model. Inference runs on this device and no image, model input, or benchmark output is uploaded.
What this AI benchmark measures
The test loads a trained convolutional neural network and repeatedly processes a normalized 224 by 224 image tensor. It measures model initialization, median inference latency, slower p90 latency, and completed inferences per second. Model download time is reported separately and does not affect the AI compute score.
WebGPU and CPU results
CPU/WebAssembly is the compatibility baseline. Chromium-based browsers may also run the model through WebGPU. The fastest completed backend determines the score, while both backend results remain visible for comparison. WebNN is detected separately because browser support is still limited.
How to get a stable result
- Close demanding games, video editors, and other benchmark tabs.
- Connect a laptop to power and disable battery saver.
- Run the test twice so model and runtime files are cached.
- Compare results using the same browser version and power mode.