Cut Opus 5.5 video QA cost 90% by routing frame checks to a cheap decision model
_AustinCalvert_ · x · 2026-10-20
Reviewing thousands of frames with Claude Opus 5.5 in Remotion burns time and tokens. @shannholmberg shares a delegation workflow: code checks measurable rules (text size, blank/frozen frames, brand colors), a cheap decision model screens the frames, and only flagged frames go to Opus 5.5 for targeted review and code fixes.
Path A has Opus review every frame; Path B sends just 2 of 6 flagged frames to Opus after cheap screening. Both paths end with re-render, recheck, and human review of pacing, audio, and transitions. Reported results: 90% lower review cost, 75% less review time — actual savings depend on check accuracy, call pricing, and frame volume.
More from coding & agent
- RL fine-tuning Qwen3.5-0.8B for under $20 on HF Jobs lifts pass@1 from 10% to 36% — LysandreJik · 2026-10-20
- DeepSeek beats Claude and Codex on Google Maps scraping speed, at a steep price — teortaxesTex · 2026-10-20
- An AI claimed it ran a calculation it never ran — 'presumption' may explain many AI errors — CourseSome8119 · 2026-10-20
- After OpenAI's rogue agent disclosure, how banks and insurers are locking down LLM agents — OkInterest1608 · 2026-10-20
- Vibe-Coded JS Landing Pages Are Breaking Meta/Google Ads Tracking, Warns Practitioner — gaganghotra_ · 2026-10-20
- Could 10,000 AI Agents Design a Fusion Reactor? Jason Crawford Thinks It's Worth Trying — burny_tech · 2026-10-20