ViewTube

benchwerk measures developer infrastructure on real hardware and publishes every number with its raw data.

Each video is one original benchmark: a prediction written before the first run, the measurements, what we got wrong, and independent reviews. The harness, the raw results and the review record are on GitHub: https://github.com/benchwerk

Shorts teach one concept each, from our own demos and data (playlist: benchwerk shorts).

How this is made: narration is synthetic speech. Scripts are drafted with AI agents and edited and approved by a human. Every measurement is run on real hardware.