ChatGPT and Codex will share a usage allowance in December – what agents need to know

OpenAI describes the transition as part of bringing Work's agentic capabilities into the main ChatGPT experience. Eligible ChatGPT and Codex usage will be measured in tokens from a shared pool. This is not a published per-task price list or evidence that all conversations count against the same limit.

Our editorial interpretation is that complex research and software tasks become part of a more unified capacity plan. Their exact effect on an individual's usage windows remains unmeasured until the rollout reaches the account.

Plus, Pro and Business are expressly included. Ordinary Instant chat is expressly excluded. This distinction matters because a user asking a short everyday question should not be told that it automatically consumes the planned agentic allowance.

The page does not define an identical December mechanism for Free, Go, Enterprise, Education or separately billed API activity. It is not responsible to infer universal inclusion or guaranteed exclusion from that silence.

A coding agent may inspect repositories, run tools and iterate through tests, while a brief chat generally has a different workload profile. OpenAI says consumption depends on model selection and task complexity. A precise conversion between a Codex run and a number of chats would therefore be invented.

Separate usage guidance discusses settings and usage windows. It can help people understand current Work and Codex allowances but does not establish a new universal December price or reset interval.

AgentenCode recommends separating Instant queries, longer knowledge work and repository or cloud coding tasks. Teams should assign priority to production-critical tasks and consider whether every request needs the most intensive model and reasoning mode.

Once the new system actually arrives, measure a small set of representative tasks against the visible Settings → Usage display and repeat the measurements consistently. Claims made before that point are hypotheses, not benchmarks.

Our seven-row matrix keeps explicit vendor statements apart from operational recommendations: rollout, plan scope, eligible work, Instant exemptions, measurement, dashboard visibility and unknown prices.

The original SVG is a source-informed explanatory model of task and model choice feeding a shared allowance. It is not a screenshot of a real account.

The announcement specifies December but not a calendar day. It does not publish one absolute token allowance for every named plan. Account-specific workspace policies, reset periods and credit mechanisms still need to be checked in official documentation at the time of rollout.

OpenAI expects most users not to wait for resets more often than before; this is a vendor expectation, not an independent longitudinal study. Nor does a shared allowance mean every external tool or API is included for free.

The important change is the convergence of coding and knowledge-work agents under a common usage model. Users should prioritize critical projects and watch remaining capacity and reset times after the transition.

Do not install untested budget rules using made-up quotas. Build a workload inventory and use real account data once the December model is genuinely available.