OpenAI allows Dots agents to operate around the clock without consuming the user’s usual usage allowance during the first month after launch, but this feature does not extend to all work performed by the agent. When a Dot routes a task to Codex or ChatGPT Work, usage begins counting according to the limits of the product that received the task.
This detail appeared in a clarification published by Thibault “Tibo” Sottiaux, head of core products and platform at OpenAI, on X, in response to a community comment questioning whether persistent agents could operate within current subscriptions without exceeding the permitted limits.
What Counts and What Does Not?
According to Sottiaux’s clarification, the user’s basic Dot is included in the plan and can operate 24 hours a day, and it does not consume the allowance when it performs the work directly. However, if the user asks the Dot to create a task in Codex, the Codex task is treated as normal usage and deducted from that product’s available limit.
OpenAI says that basic Dots usage will not count toward eligible allowances during the first month after launch, but it will publish the usage terms for each plan after that period ends. This differs from Sottiaux’s wording, which described the core functionality as remaining included in the plan; therefore, developers still do not know how much usage will be available to them in the future.
Why Does the Task’s Execution Path Matter?
A Dot may begin working within the included functionality and then decide to send part of the work to another product with separate limits. In practice, the impact of a task depends not only on its nature but also on where the agent executes it. Routing a programming task to Codex may consume the allowance even if the Dot itself remains available without an additional usage charge.
This issue is more important for developers planning to build continuous automation around Dots, as they will have to monitor routing between products in addition to choosing a model and managing a token budget. The source did not clarify whether users will receive an alert before a task moves from included usage into metered usage, leaving open the possibility that they may discover the allowance has been consumed only after it happens.
A Paid Tier and Lower Capacity for Some Plans
OpenAI expects the number of active Dots to reach a few million within days, and the company has absorbed their core activity during the launch period. However, it has not yet disclosed what future limits will look like for each plan, at a time when it also cut the allowance included in the $200 Pro plan in half on the same day and added a higher tier priced at $500.
Sottiaux also hinted at a paid tier that would increase a Dot’s speed and give it greater “bandwidth,” while keeping the core functionality within the plan. This means the expected pricing model may separate the agent’s existence and basic operation, on the one hand, from execution speed or workload volume, on the other.
certi.news’s Take
The real change here is not merely the launch of an agent that operates continuously, but the shift of cost from the agent’s existence to the task path it chooses. Since a Dot may decide to delegate work to another product, clarity around routing rules, alerts, and the limits of each plan will be crucial for users who rely on it for long-running or repetitive tasks. Until OpenAI publishes the detailed terms after the first month, the future operating cost and actual usage limits remain two open questions.