Codex bug double-counts historical reasoning tokens, triggering premature auto-compaction at ~55% usage
KiwiGeorge · ghdev · 2026-10-12
A bug report against Codex App 0.162.0-alpha.17.2 claims the client re-estimates historical reasoning tokens already counted server-side, inflating the auto-compaction counter. A Pro Max $500 user with a 1M context window and 900k threshold saw compaction start at 55% displayed usage (459,164 tokens): 459,164 server-reported + 324,498 estimated historical reasoning + 4,737 pending tool tokens = 788,399, crossing the effective 784,800 threshold. Compaction took 199 seconds. A minimal repro with gpt-6.1-sol at high effort shows an extra 244–268 tokens on both SSE and WebSocket, with server per-item attribution confirming the reasoning item was already counted.
More from coding & agent
- Garry Tan endorses Aside Browser as the fix for agents getting blocked on the web — garrytan · 2026-10-23
- 1Password launches tool letting AI agents log in for you without seeing your password — AndrewSchmidtFC · 2026-10-23
- MCP server turned rogue 30 days after approval, prompting open-source mcp-guard lockfile tool — ErgoniaWorks · 2026-10-23
- Test: Coding Agents Won't Use an MCP Tool Unless AGENTS.md Mentions It — colinmcnamara · 2026-10-23
- Open-Source Tennis Models Plus Claude Turn Video Into a World-Class Coach — tweetsatpreet · 2026-10-23
- awesome-ai-apps: 132 Free Open-Source AI Agent and RAG Projects on GitHub — Arindam_1729 · 2026-10-23