OpenAI announced a planned shared usage allowance for ChatGPT and Codex.
The change can affect how people budget interactive work, but its exact mechanics remain unpublished.
What happened?
OpenAI says that starting in December, Plus, Pro and Business plans' usage allowance will be shared across ChatGPT and Codex. The notice says usage will be measured in tokens rather than a single universal message count, and that more advanced models and complex tasks generally consume more allowance. Combined remaining usage and reset times are expected to appear in Settings → Usage.
The notice also names an important exemption: Instant chat will not draw from the shared usage. Users who hit the relevant limit can wait for a reset, upgrade or buy credits for eligible features. OpenAI says it expects most users will not need to wait for resets more often than they do now, but that is an expectation rather than a service-level guarantee.
OpenAI's notice does not promise that a shared counter will preserve the current number of messages for any plan. It says more advanced models and complex tasks generally use more of the allowance, so the eventual user experience will depend on the model mix and task shape as well as the plan name. That makes the planned Settings display important for practical budgeting.
What the notice does not say
OpenAI has not published an exact December start date or per-plan token amounts in the notice reviewed. It also does not say that every ChatGPT action is merged into one pool. Instant chat is explicitly excluded, while treatment of other model classes, Work, Excel, PowerPoint, Enterprise/Edu, Free/Go and credits remains feature- and plan-specific in the surrounding documentation.
This is a subscription-account change, not a general API price or capacity announcement. API requests follow their own billing and rate-limit documentation. Likewise, a credit purchase for an eligible ChatGPT feature should not be described as an API credit or as a permanent increase to a plan's allowance.
The notice's token language also changes what a user can infer from a visible task. A short prompt may invoke a costly model or a long agentic workflow, while a complex request may consume more than a comparable chat exchange. The surrounding Codex help explains that plan usage is distinct from API billing and that feature eligibility can vary by plan. The credits page describes credits as a separate way to extend certain eligible personal-plan features, not as a universal replacement for the plan allowance. Until OpenAI publishes the table, readers should not translate the announcement into a fixed number of prompts per day.
Source: OpenAI ↗ · Source: OpenAI ↗ · Source: OpenAI ↗ · Source: OpenAI ↗
Why does it matter?
Sharing the allowance changes the user's optimization problem. A heavy Codex session can consume from the same plan-level budget that a person expects to reserve for ChatGPT work, even if the visible interfaces remain separate. Token accounting may be more faithful to compute cost than message counts, but it is also harder to predict before a task runs. Teams should watch the Settings usage display, record which model and mode a workflow uses, and avoid budgeting against today's message behavior until OpenAI publishes the token schedule.
The biggest risk is false precision. “Shared” does not tell a user how many coding tasks remain, whether a particular feature is excluded, or whether buying credits helps with the exact limit they reached. The useful headline is therefore a boundary change with missing mechanics, not a promised reduction or increase in everyone's capacity. The documentation should be read as a rollout notice and revisited when the December date and plan tables arrive.
For a team, the change increases the value of usage observability. A shared pool can simplify billing and reduce the need to reason about two separate counters, but it can also make one person's heavy coding workflow visible as a capacity problem for another person's chat work. Product owners should define a fallback path, watch reset behavior during the rollout and document which tasks are allowed to consume purchased credits. Those are sensible controls while the provider has announced the accounting model but not the final numbers.
Source: OpenAI ↗ · Source: OpenAI ↗ · Source: OpenAI ↗ · Source: OpenAI ↗
