OpenAI Pauses 5-Hour GPT-5.6 Sol Cap as Codex Demand Spikes

OpenAI has removed the five-hour usage window for GPT-5.6 Sol across ChatGPT Plus, Pro, and Business accounts, granting developers and IT teams a temporary reprieve to run longer coding sessions and agentic workflows. The change, confirmed by product manager Tibo through a report on AIBase, arrived after a 48-hour surge in demand for Codex and ChatGPT Work strained the model’s capacity.

A curtain lifted—but not removed

The most visible shift: users will no longer see a hard five-hour timeout before needing a reset. OpenAI also issued a one-time reset of the available agentic-workflow quota, giving current subscribers a fresh start. But the company has not published a written notice specifying how long this relaxed policy will last, leaving teams to assume it is temporary.

The official GPT-5.6 Sol documentation still warns that usage depends on the plan and, for managed workspaces, local settings. Lifting the visible timer does not mean unlimited use. Paid users may still hit a total plan allowance, credit requirements, model fallback to a less capable version, or other service-side throttles. The core rule remains: limits exist to manage capacity and prevent abuse.

Who benefits most—and who should stay cautious

The loosened cap matters most to developers and power users pushing Codex, ChatGPT Work, ChatGPT for Excel, or Workspace Agents through extended tasks. These agentic products draw from the same usage and credit pool when included in a plan. A long-running repository analysis, an afternoon-long debugging session, or a cloud-delegated workflow can eat through quota much faster than plain conversation.

For Windows developers, the immediate effect is more headroom to finish complex jobs without hitting the timeout. The ChatGPT desktop app’s Codex mode becomes more viable for marathon sessions, provided you’re on the right version. Admins managing Business or Enterprise workspaces should brief their teams: the cap is paused, not retired. Role-based access and credit policies still apply, and plan admins can adjust local limits from the workspace dashboard. If your organization has set a lower custom cap for Sol, that restriction remains unless you change it.

For everyday subscribers, the change is less dramatic. Casual users who rely on GPT-5.6 Sol through the Medium or High reasoning settings will notice fewer interruptions, but the lift mainly eases the friction for power workflows. If you never hit the five-hour mark, this news buys you nothing—yet.

The efficiency drive behind the scenes

While pausing the time limit, OpenAI is also retooling GPT-5.6 Sol to consume fewer tokens for comparable work. Tibo confirmed the efficiency push, though the technical details remain private. Industry speculation points to lighter architecture and improved token generation, but no engineer has put a name to the method.

OpenAI’s own preview materials support the efficiency narrative. The company claims Sol delivered stronger GeneBench results than GPT-5.5 while using fewer tokens. On one cybersecurity benchmark, it matched earlier performance with roughly one-third of the output tokens. Those are controlled test results, not guarantees for a specific codebase or workflow. In practice, token consumption will still depend on repository size, tool calls, reasoning depth, and whether you pick Medium, High, or Extra High.

The dual move—lift the cap temporarily and make the model cheaper to run—suggests OpenAI is trying to soften the impact of heavy usage spikes without permanently sacrificing capacity control. For users, the real long-term prize is efficiency, not a free pass.

How we got to a five-hour limit

GPT-5.6 Sol arrived with time-based caps baked in. OpenAI has long used a mix of rate limits, token windows, and model fallbacks to juggle server load across millions of accounts. Early ChatGPT models had per-minute request ceilings; later versions added hourly quotas. The five-hour limit on Sol was a relatively aggressive guardrail, likely born from anticipating the resource hunger of deep reasoning and coding workflows.

The recent spike came from Codex adoption. As more developers plugged the model into CI/CD pipelines, Excel macros, and agentic loops, the workload outstripped what the original cap could smoothly handle. Rather than risk developer frustration—or lost subscription revenue—OpenAI chose to pause the limit and reset quotas.

This isn’t the first time the company has adjusted caps on the fly. GPT-4’s message limits expanded and contracted several times after launch, always without a public schedule. The pattern: loosen when demand peaks, tighten again when the rush subsides or efficiency catches up.

What to do right now on Windows

If you’re a Windows user leveraging GPT-5.6 Sol in the desktop app, a few concrete steps will help you make the most of the temporary opening—and avoid surprises when limits return.

1. Update the desktop app. Codex mode for GPT-5.6 requires ChatGPT desktop app version 26.707.30751 or later. Check for updates from the app’s Settings and confirm your version. If you’re behind, the new model might not appear.
2. Verify Codex CLI. If you use the command-line tool, minimum version 0.144.0 is needed for GPT-5.6. Run codex --version and upgrade if necessary.
3. Monitor your plan’s usage page. OpenAI shows remaining allowances and credit consumption in your account settings. The cap lift does not remove plan-level maximums, so keep an eye on those meters during long sessions.
4. Adjust workspace settings if you’re an admin. For Business or Enterprise plans, log into the admin console and review the model availability, role permissions, and any local credit caps you may have imposed. Communicate to your team that the lift is temporary; plan accordingly for limits to snap back.
5. Expect fallback behavior. If you exceed the total allowance or credit pool, Sol may fall back to a less capable model (like Terra) or suspend agentic actions. Bookmark the official documentation for your plan’s current limits.

The table below summarizes the model breakdown for different plans, as listed in OpenAI’s support materials.

What comes next

OpenAI has not given a deadline for reinstating the five-hour cap, and its silence on the timeline is deliberate. The company will likely watch usage patterns, capacity headroom, and how quickly the efficiency improvements roll out. If the model truly starts burning fewer tokens per task, the old cap might return in a loosened form—or not at all.

For Windows users, the wise move is to treat this as a window, not a new normal. Keep your tools updated, budget for limits to return, and take advantage of the extra headroom while it lasts. The bigger story, efficiency, will play out over months and could reshape how much work you get from a single subscription.