Article

Codex "Model Full" errors have been significantly reduced: the proportion of affected users has dropped from 6.82% to 0.085%

Dynamic Observation Beating AI news flash: on OpenAI, core product lead Tibo Sottiaux shared an internal statistic showing that the team may have finally resolved the long-standing "model capacity full" error plaguing users.

The data shows that on October 19, about 6.82% of Codex daily active users encountered at least one request that ended with a capacity error. By November 12, this proportion had fallen to 0.085%.

Previously, many users reported that even with sufficient subscription credits, Codex would suddenly prompt "model full," requiring them to switch models. Ongoing programming tasks would be abruptly interrupted, and switching models didn't always help.

This fix primarily focused on the automatic retry mechanism. October 15 to November 6,OpenAI made 6 rounds of adjustments in succession. When the server is overloaded, it tells Codex how long to wait before retrying. After receiving the instruction, the client can wait and automatically retry, minimizing direct task termination. OpenAI also increased inference compute power beforehand.

Some joked that the reduction in Codex errors might be because users have all switched to using Claude Opus 5.5.