One AI video job burned 56M tokens, mostly re-reading the same cached context
dotey · x · 2026-10-03
dotey shares the token bill for an AI video generation job: 56M tokens total, but only 180K were actually generated — nearly all the rest was the same context re-read from cache. He notes the cost was inflated by iterating with many revision requests; using his final prompt would need only 1-2 runs, and a subscription plan would be cheaper still.
More from Apps
- How an AI-made Bach podcast was built with Gemini 3.8 Flash and heavy human steering — sytelus · 2026-10-03
- Dev uses AI to build a free podcast that unpacks Bach measure by measure — sytelus · 2026-10-03
- Chrome's Hidden Cmd+Shift+A Shortcut Lets You Instantly Search Your Tabs — manorlaboratory · 2026-10-03
- Grok Bot, Meta Muse, Dot, Hermes: which AI agent to use for which task — HarperSCarroll · 2026-10-03
- Parent uses ChatGPT to generate ten graded polygon exercises in seconds — charitychaste · 2026-10-03
- Solo founders' week: 164MB speech model beats Whisper Large, $117M raised for AI datacenter turbines — julianweisser · 2026-10-03