OpenAI Restores 5-Hour Limit for Plus Users of Codex
Thibault Sottiaux, head of OpenAI Codex, announced that Plus users will have the 5-hour usage limit re-enabled the following day, affecting both ChatGPT Work and Codex.
This limit was previously suspended. Sottiaux explained that the restoration is mainly due to two reasons: first, to spread out computing demand and avoid a large number of users using the service at peak times, thus maintaining a generous weekly quota; second, to prevent Plus users (mostly casual or new users) from feeling confused after exhausting their weekly quota all at once.
The $100 and $200 Pro tiers will continue to not have the 5-hour limit for at least the next few months. After reaching the limit, users can either wait for a reset or purchase additional credits, and OpenAI occasionally offers free resets.
Enterprise and educational accounts currently use a separate credit system, with no changes mentioned. During the previous temporary suspension, only the weekly limit was retained.
Driven by events, OpenAI manages computing power allocation through tiered limits, prioritizing funding and resources to ensure a continuous experience for high-paying Pro users. The beneficiaries are Pro subscribers and overall system stability, while Plus users face short-term workflow pressures.
Source: Public Information
ABAB AI Insight
Thibault Sottiaux, as the head of OpenAI Codex and ChatGPT engineering, has publicly adjusted usage limits multiple times to respond to changes in model efficiency and user behavior. Previously, due to the more aggressive usage patterns of the GPT-5 series models, the 5-hour window was temporarily canceled and quotas were reset; now, the restoration for Plus tier aims to balance the load.
On the capital path, OpenAI spreads computing costs through differentiated limits, with Plus as the entry-level paid tier bearing more short-term constraints, while the Pro tier retains flexibility to maintain high-value user retention. The motivation is to protect the generosity of the weekly quota during times of tight computing resources, while reducing support costs and churn for new users due to accidental quota exhaustion.
This is similar to the multiple adjustments in rate limits after the early launch of ChatGPT Plus, or the tiered quota strategies of competitors like Claude; currently, it is in a phase of refined resource management under parallel multi-model and high usage intensity.
Essentially, this represents a transfer of pricing power. As computing power becomes a scarce resource, the platform redefines user rights through time windows and tier differences, with the mechanism smoothing peak demand with short limits, shifting cost and experience pressure towards lower-paying tiers, while locking in the continuous productivity advantage of higher-paying tiers.
ABAB News · Cognitive Laws
Peak computing power determines product rules
Entry-level tier bears fluctuations, high-end tier locks in stability
Temporary cancellations will eventually return due to load