Power user tallies 12B coding-agent tokens, finds 96% were cache reads
peakpirate007 · reddit · 2026-11-04
A Reddit user auditing their coding-tool usage found 12B+ tokens (Claude Code 5.4B, Codex 6.7B+), but session logs showed 96% were cache reads — the same context re-read across turns. They also found old Codex sessions were priced at their current service tier since tiers weren't recorded, inflating cost estimates. They open-sourced StackPulse, a local-first CLI supporting 13 coding tools that generates HTML reports splitting input, output, cache reads and costs (npx stackpulse).
More from coding & agent
- IBM open-sources CUGA, an enterprise generalist agent framework with MCP support — samuelcolvin · 2026-11-05
- Does more model intelligence still matter for coding past a threshold? — the_juan_1 · 2026-11-05
- Get a real second opinion: send the failure, not the agent's theory — Ok_Negotiation_2587 · 2026-11-05
- Dev builds code-driven Poison Ivy scrolling component in Framer with multiple configs — evilrabbit_ · 2026-11-05
- The Missing Primitive in AI-Employee Tooling: Agents Can't Watch and Wait — corbtt · 2026-11-05
- rustgo: a from-scratch Rust compiler written in Go, mirroring rustc's pipeline — rakyll · 2026-11-05