Anthropic shares 3-layer playbook to cut Claude API costs without losing quality
xiaohu · x · 2026-09-09
Anthropic's developer team published guidance on cutting Claude API bills without degrading output: before switching to cheaper models, audit three layers where you may be paying for zero-value work. Input layer: re-prefill of repeated content — cache reads require byte-exact prefixes, are model-bound, and the default 5-minute TTL breaks on long agent tool calls; mid-conversation Effort changes, dynamic variables like timestamps, and tool definition reordering all bust the prefix. Instruction layer: legacy verification rituals, forced steps, stale examples, and conflicting rules make frontier models do unnecessary work. Reasoning layer: calibrate Effort per task difficulty by sweeping Effort/model combos with your own evals — too high wastes tokens, too low hurts quality. Official tools prompt-audit, cost-optimize, and hillclimb cover prompt cleanup, cost auditing, and configuration search.
More from coding & agent
- SF 'Code Engineering' event picks up where 1,000-person harness engineering meetup left off — LangChain · 2026-10-27
- Prompt turns Claude Code into an iPhone backup forensics tool exposing 30-day 'deleted' photo copies — alex_verem · 2026-10-27
- Prompt turns Claude Code into a forensics tool finding deleted iPhone photos in backups — alex_verem · 2026-10-27
- Phones won't be mobile IDEs but control planes for agent fleets, argues Yuchen Jin — Yuchenj_UW · 2026-10-27
- FrontierCS 2 launches on Kaggle: 48 tasks make agents beat published CS papers — MeganRisdal · 2026-10-27
- Stop remaking old games with AI: a call for more ambitious agentic software — rschu · 2026-10-27