OpenAI is removing the frustrating 5-hour usage windows for its new GPT-5.6 models, allowing paid ChatGPT users to burn through their weekly limits at their own pace. Following a massive surge to 6 million active users, the company is taking a starkly different approach from Google's heavily restricted Gemini 3.5 launch by optimizing inference and easing bottlenecks rather than locking users out.
In a post on X, Thibault Sottiaux, OpenAI’s newly appointed head of core products, announced the temporary removal of the 5-hour usage limit restriction exclusively for Plus, Business, and Pro users. This shift means power users can now exhaust their entire weekly allocation in a single marathon session if a project demands it.
Addressing community concerns about potential performance throttling, Sottiaux noted in a separate post that the team is actively experimenting with different reasoning efforts, particularly in multi-agent scenarios. While the reasoning capability was briefly reduced as part of an internal test, it has since been reverted to its original baseline.
Technical Adjustments for Unrestricted Access
To balance the removal of time-based throttles without melting their infrastructure, OpenAI has implemented a few strategic backend adjustments:
- The context window limit has been shrunk back to 272,000 tokens, down from the previous 372,000, to reduce overall compute load per prompt.
- New inference optimizations have been deployed, resulting in a 10% increase in total usage capacity specifically for the GPT-5.6 Sol model.
The GPT-5.6 Family and ChatGPT Work
The GPT-5.6 family recently launched after a brief preview phase, splitting into three distinct tiers: Sol, Terra, and Luna. GPT-5.6 Sol serves as the heavy hitter, designed for demanding chores like complex coding, creative design, and deep data analysis. On the other end, Luna is optimized for cost-efficient tasks, though OpenAI claims it still outperforms Anthropic’s Claude Opus 4.8 and 5. Terra sits in the middle, offering a balance of speed and reasoning.
Alongside the core models, OpenAI rolled out ChatGPT Work, a direct competitor to Claude Cowork. The new workspace environment gathers context across desktop apps and web browsers to coordinate professional tasks autonomously. To streamline this ecosystem, OpenAI has also upgraded its standalone Codex app and fully merged it into the ChatGPT Work interface.
The Strategy Behind the Token Squeeze
Dropping the context window from 372k to 272k tokens is a calculated trade-off. OpenAI is sacrificing massive document ingestion capabilities to keep the servers responsive for high-frequency, shorter queries. By removing the 5-hour throttle, the company is betting that power users prefer sprint-bursts of productivity over forced pacing.
This move also highlights a stark contrast in infrastructure confidence. While Google capped Gemini 3.5 heavily at launch to manage compute costs, OpenAI is leaning into inference optimizations to squeeze out 10% more efficiency. The integration of the Codex app into ChatGPT Work suggests OpenAI isn't just trying to win the model benchmark war - they are aggressively building an inescapable desktop OS layer for enterprise users.
