MicroLLM Lab is a web-based platform that allows users to run and benchmark seven different tiny Large Language Models (LLMs) directly within a web browser. The tool utilizes the WebGPU engine to execute models and measures performance through objective checks, such as regex and exact token matching, rather than qualitative writing assessment.
MicroLLM Lab enables benchmarking of seven tiny LLMs in the browser via WebGPU
The platform provides several performance metrics, including speed in tokens per second (sustained decode and suite wall) and accuracy based on pass rates from objective tests. Performance data, such as token generation speed, is stored locally on the user's machine using IndexedDB cache. Users can also generate and download a verifiable performance certificate that includes their device hardware specifications as well as peak and sustained tokens per second.
Sources
- MicroLLM Lab – Try 7 tiny LLM's in the browser （Hacker News Frontpage, 2026-09-28）