Codex Pro User Reports 30 Hours of "Model at Capacity" Errors Burning 35% of Weekly Usage

EzzaGames · ghdev · 2026-09-21

Issue #46980 on openai/codex: a ChatGPT $200 Pro (20x) subscriber reports 30 hours of continuous 'model at capacity' errors across Astra and Luna, with working Astra responses degraded and needing babysitting — burning 35% of weekly usage on wasted tokens. User demands a fix and banked resets as compensation.

Original post →

More from Models

Models channel →