24-Hour Agent Run: 98.5% Cached Input, Reasoning Eats 72% of Output Tokens
Clear_Evidence9218 · reddit · 2026-10-23
A developer shared token statistics from long-running autonomous agents and asked for community benchmarks. A 24-hour agent run totaled 22.78M tokens: 16.7M uncached input, 1.09B cached reads, 6.07M output (4.38M of it reasoning).
Key ratios: input-to-output 2.75:1 (excluding cache); output is 36.4% of uncached input; cached input is 98.5% of all reported input tokens; reasoning is 72% of output. A compiler-building agent running 6+ days showed remarkably similar ratios.
The author notes tokens aren't productivity, but finds the stability of the fresh-input/cached-context/output relationship across long autonomous runs notable—and suggests architectural decisions in agent environments can meaningfully shift these numbers.
More from coding & agent
- Dev previews proactive open-source AI agent: works with any provider or harness, zero-config — max_paperclips · 2026-10-23
- Dev shares two AGENTS.md lines to make AI-generated UIs look less like AI — pjausovec · 2026-10-23
- Meta, Google, OpenAI and xAI all converge on cloud computers for personal agents — sujingshen · 2026-10-23
- diffy semantic diff app to auto-discover Git worktrees in next release — seatedro · 2026-10-23
- How a 5-person AI film crew produced an Odyssey pilot scene in Imagine — chaitu · 2026-10-23
- "One prompt built it all!" Frontier dev: we refactor like this several times a day — fkasummer · 2026-10-23