WebLLM Benchmark

Best experienced on Chrome or browsers with WebGPU support

SELECT EXECUTION BACKEND

⚠️ WebGPU unsupported — Locked to WASM CPU Fallback mode

CHOOSE AI MODEL

Detecting GPU...

4 testsUp to 704 total tokens4 iterations each

Speed Test128 tokens

Basic generation speed

Math Reasoning256 tokens

Simple arithmetic

Logic256 tokens

Simple deduction

Creative Short64 tokens

Short creative writing

How to Run a Valid Benchmark

Follow each step precisely. Skipping stages can invalidate your submission or skew leaderboard rankings.

01
Use a Supported Browser
Run benchmarks in Google Chrome, Microsoft Edge, or another Chromium-based browser. Set GPU preference to 'High performance' before running tests.
02
WebGL & WebGPU Support
WebGLWebGPU
Both WebGL and WebGPU must be hardware accelerated. Visit chrome://gpu and confirm both show 'Hardware accelerated'.
03
Enable High Performance Mode
Enable 'Use hardware acceleration when available' in browser settings. Switch to High Performance power plan to prevent throttling.
04
Close Background Tabs & Apps
Close unnecessary tabs and background apps. Memory pressure will skew inference speed and efficiency scores.
05
Run Each Test Twice
Perform a warm-up run before recording results. Submit only the second run.

All submitted scores are verified server-side. Non-compliant submissions may be excluded from the leaderboard.

LiteEdgeAI evaluates hardware using real-world AI workloads, focusing on inference speed, memory efficiency, and sustained performance.

Every score is fair, explainable, and repeatable — so rankings actually mean something.