Codex 'godmode' config: 1M context window, low reasoning effort and a 400K auto-compact limit

EXM7777 · x · 2026-09-06

A developer shared a GPT-Astra settings set for Codex that reportedly unlocks 'godmode': set modelreasoningeffort to low, modelcontextwindow to 1,000,000, modelautocompacttokenlimit to 400,000, and leave the global AGENTS.md empty. The idea is a huge context window with a high auto-compact threshold to avoid mid-task compaction, while low reasoning effort keeps responses fast. Results will vary by task.

Original post →

More from coding & agent

coding & agent channel →