OpenAI is temporarily removing the five-hour rolling usage limit on GPT-5.6 Sol for Plus, Pro, and Business plan subscribers after 48 hours of surging demand from its Codex coding agent and ChatGPT Work autonomous task manager. The company also issued a one-time usage reset and announced efficiency improvements to the model.
What Changed
OpenAI product lead Tibo confirmed the changes in a post on X on July 12. “The last 48 hours of Codex and ChatGPT Work have been intense,” he wrote, announcing three updates: the temporary removal of the five-hour usage restriction, efficiency changes to GPT-5.6 Sol that “will be reflected in less usage being used so that it can take you further,” and a one-time usage reset.
The post also disclosed that GPT-5.6 Sol has reached 6 million active users, with 2.8 million views on the announcement within hours.
Under normal operation, ChatGPT and Codex count both local messages and cloud-based tasks against a shared usage limit using a rolling five-hour window, with weekly limits varying by plan and model. The temporary change removes the five-hour cap, so users are no longer forced to stop working when they exhaust that window. Weekly limits may still apply, according to BleepingComputer’s reporting.
Infrastructure Strain
The move reveals a capacity constraint that frontier AI labs increasingly face when agentic products drive sustained compute demand. Codex and ChatGPT Work are both designed for autonomous, multi-step work: Codex runs coding tasks in cloud sandboxes, while ChatGPT Work handles email, Slack, and calendar operations without user intervention. Both consume significantly more compute per session than conversational ChatGPT use.
When demand for these agentic features spiked over two days, OpenAI’s provisioned inference capacity was not enough to serve all users within their normal rate limits. Rather than degrading the experience, the company chose to temporarily lift restrictions and push efficiency improvements to reduce per-request compute cost.
The efficiency changes suggest OpenAI is working on the model’s inference pathway to reduce token consumption or optimize compute allocation. Tibo said the exact impact would be “quantified and shared” but did not provide a timeline.
Precedent and Pattern
This is not the first time a frontier lab has adjusted usage limits under demand pressure. Anthropic extended free access to Claude Fable 5 three times (from July 7 to July 12, then to July 19) as it worked through its own pricing transition. The pattern points to a recurring tension in agentic AI: autonomous agents consume far more compute than chat-style interactions, and the infrastructure to serve them at scale is not yet sized for peak demand.
For teams relying on GPT-5.6 Sol for production agentic workflows, the temporary relaxation provides breathing room. The longer question is whether OpenAI’s efficiency improvements will be sufficient to avoid future capacity crunches as Codex and ChatGPT Work adoption continues to scale.