One Claude session generated a full video explaining Attention, burning 271K context tokens

Abhishekcur · x · 2026-09-29

Abhishekcur ran a stress test: he asked Claude (referred to as Sonnet 5.5, Effort set to Medium) to create a video explaining Attention from scratch. The session consumed 271.8K of the 1M-token context window, as he pushed to see what the model could actually do in a single long-context, end-to-end content production task.

Original post →

More from Fun

Fun channel →