What a Claude Code task really costs on Opus 5.5: cache hit rate is everything
xiaohu · x · 2026-09-23
A breakdown based on Addy Osmani's post at Anthropic. Opus 5.5 cut API prices across the board: fresh input $4/M (-20%), output $20/M (-20%), cache reads $0.20/M (-60%, just 5% of fresh input).
For a representative 40-turn task averaging 70K tokens per turn (2.8M cumulative input), paying all fresh input would cost $11.20; at a 90% cache hit rate input drops to $1.62. The key insight: Claude Code's cost blowup comes from re-sending history every turn, so maximizing prompt cache hits is the biggest lever, followed by fewer turns and less output (thinking is billed as output). Also covers effort tuning, model choice and /usage, with two reusable templates.
More from coding & agent
- Matt Pocock: Stop chasing model releases, improve your agent's harness instead — mattpocockuk · 2026-09-23
- Uncle Bob: AI changes nothing—complexity, not tooling, still makes software slow — blaizedsouza · 2026-09-23
- GBrain: plug your own memory, tools, and skills into any AI — garrytan · 2026-09-23
- Podcast: building a playable game with $8 of parts and AI assistance — aishashok14 · 2026-09-23
- Agent kept searching but never opened the source: four runs expose a hidden failure mode — memokris · 2026-09-23
- TypeSafe classifies RAG passages with probability thresholds to fight noise and prompt injection — marlene_zw · 2026-09-23