OpenAI adds compute to fix Codex capacity errors, tasks to auto-resume

OpenAI's Codex and ChatGPT lead announced on social media that the team has rolled out a series of changes and added more computing power to fix the "model at capacity limit" error message. This update is part of day 11 of their 28-day continuous delivery plan, aiming to make this error much less frequent.

Public discussions suggest the issue isn't just models being truly full. Some users reported the error kept appearing even with plenty of subscription quota left and after switching to different models. Some developers suspect it's linked to backend requests, session states, or resource scheduling.

To fix this, OpenAI optimized its systems and added more compute to reduce the errors. They also improved task recovery, so tasks can now automatically continue after an error instead of getting stuck and waiting for users. The team is asking users who still see the problem to share feedback for further checks.