Codex Announces Rate Limit Fix and Reset
Codex head Tibo posted an update on the rate limit issue. The team found efficiency problems with image compression during long sessions, high p95+ usage in Computer History, and the generation of conversation titles consuming more than expected.
The team is conducting a thorough investigation and plans to release a fix tomorrow. They also discovered a new method for efficiency improvement, which will be advanced next week.
As part of tomorrow's fix, a complete usage reset will be applied to all paid subscriptions. The reset is expected to take effect at 2 PM Pacific Time tomorrow.
Source: Public Information
ABAB AI Insight
As an OpenAI-related coding tool, Codex has previously managed computing resources through a subscription model, facing long-term usage fluctuations due to long sessions and multimodal inputs. Tibo's team publicly diagnosed specific bottlenecks, indicating that internal monitoring of resource consumption has covered image compression and historical functions.
In terms of capital strategy, resources are shifting from temporary limits to systematic fixes and full resets, motivated by the need to maintain paid user retention through immediate compensation, while paving the way for next week's efficiency plan to reduce long-term server costs and support higher concurrency.
Similar cases can be seen with models like ChatGPT and Claude, which alleviated user dissatisfaction through resets or expansions after early rate limit adjustments. The current phase is a transition from functional expansion to resource efficiency optimization.
The structural judgment indicates a technological substitution: inefficient functions are replaced by targeted fixes, with the mechanism prioritizing resets and optimizations after identifying high consumption paths, shifting computing resources from waste to sustainable supply, and accelerating the stability improvement of paid products.
ABAB News · Cognitive Laws
Limit exposure of bottlenecks, reset to gain trust
High consumption functions are streamlined first
Efficiency improvements often come from unexpected paths