Claude Code burns about 15–20% of tokens on 3‑word status line
Flamerix AI reports that Claude Code sends a background‑agent summary request roughly every 30 seconds. The request copies the entire transcript to the agent’s model (usually Opus) and returns a three‑to‑five‑word status line. Because the request is billed like any other Claude Code call, it can consume around 15–20% of a user’s token budget, reducing rate‑limit headroom.
Key points
- Claude Code sends a summary request every ~30 seconds, using the full transcript and costing ~15–20% of token budget.
- Independent measurements report 18–26% overhead; weekly spend share rose to ~16% by mid‑October.
- A Bun preload script (no‑agent‑summary.js) can suppress the timer, eliminating the extra token usage.
Independent measurements collected by 86879 show an overhead of 18–26 %. The author’s own telemetry on version 2.1.301 recorded a weekly spend share that grew from 3.5–7 % in July to 7.3 % then 11.8 % the following week, reaching about 16 % by mid‑October. The cost is mostly cached reads, but the uncached 2–3 % of calls carry a third of the expense.
A workaround is provided via a Bun preload script (no‑agent‑summary.js) that intercepts the timer callback and disables the summary request. Installing the script through the BUNOPTIONS environment variable (or a wrapper) stops the extra calls without altering the binary, restoring normal token usage.
The story so far
2 episodes →- Claude Code burns about 15–20% of tokens on 3‑word status linethis story
Claude Code burns ~15–20% of your tokens on a 3-word status line
gist.github.com · 29 October 2026
Loading the full article…
This text was published by gist.github.com and written by 262588213843476. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Hacker News discussion · 88 pointsnews.ycombinator.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Google plans consumer agentic AI in Gemini app by year-end · 1 src
- Sierra adds compliance tools to Horizon Agent Platform · 1 src
- OpenAI launches Decisions API for fast, structured AI judgments · 1 src
- Supabase adds agent-friendly features for AI-driven app development · 1 src
- Anthropic launches Claude Duet, a proactive AI assistant for US paid users · 2 src
Comments
via GitHub Discussions