34-day agent run processed 21.5B input tokens for $200 with 98% cache hits
nodo48 · reddit · 2026-09-10
A developer detailed a month-plus autonomous agent run: 34 days 18 hours of governed execution across two goals, mostly on GPT-5.6 Sol (extra high) then Astra (high), consuming 21.52B input tokens for $200 — about $0.00929 per million input tokens. The key was 98.25% input cache hits from reusing instructions, tools, and docs. He describes a fail-closed architecture: explicit acceptance criteria, context recovery, permissioned execution, mandatory verification before acceptance, and reopening tasks without sufficient evidence. He argues effective cost, goal continuity, and verified output should be measured together.
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
- Agent Field Kit launches $29 security runbooks for production AI agents — alvelda · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27