How ChatGPT Work usage is metered
Work does not have its own allowance. It follows the same usage structure as Codex and pulls from the same shared pool — which is why a heavy Work week quietly shrinks what Codex can do.
VERIFIED 2026-08-23 · OPENAI HELP CENTERThe one rule that explains the rest
Work follows the same usage structure as Codex. OpenAI points to the Codex pricing page for included usage and additional credits, while noting that its examples are based on coding tasks — so Work usage will vary by task. Tasks started through Voice draw from that same shared agentic usage and credit pool.
Practically: there is one agentic budget on your account, and Work, Codex and Voice-started tasks all spend from it. A four-hour Work task that builds a financial model is not free just because you were not writing code.
What consumes more
OpenAI has not published a per-task price list for Work. What it has said is that usage varies by task, which lines up with what the shape of a task suggests:
| Task shape | Relative cost | Why |
|---|---|---|
| Reformat a document you supply | Low | One source, no research, single pass. |
| Summarize a folder into a brief | Medium | Multiple files read, one artifact produced. |
| Research a market and build a deck | High | Browsing, many steps, revision loops, a formatted output. |
| A scheduled task that monitors and reports | High over time | Cost recurs on every run, whether or not anything changed. |
Scheduled tasks are the sneaky one. A daily monitor is a task that spends every single day. Set the cadence to the slowest one that still answers your question.
Practical ways to spend less
- Draft the brief in Chat, run it in Work. Thinking is cheap in Chat. Do not burn agentic budget deciding what you want.
- Hand it the sources. Attaching the three documents it needs is far cheaper than asking it to go find them.
- Say what "done" looks like. "A 10-slide deck, one chart per slide, finance audience" avoids the revision loop that costs the most.
- Scope local work to one folder. On desktop, grant access only to the files the task needs — narrower context, less spend, smaller blast radius.
- Audit your scheduled tasks monthly. Recurring runs are the line item people forget they authorized.
What OpenAI has not published
Being straight about the gaps, because most write-ups on this topic are not:
No Work-specific credit table exists. Numbers circulating for "credits per Work task" are extrapolated from Codex or from ChatGPT for PowerPoint examples, not published for Work. Treat them as estimates.
No published cap on concurrent Work tasks by plan. Per-plan numbers for active scheduled tasks appear in secondary coverage but we have not been able to confirm them against OpenAI documentation. We are not going to repeat them as fact.
When OpenAI publishes either, this page changes and the badge date at the top moves with it.