OpenAI’s agents hit 10 million users as ChatGPT Work push accelerates

The milestone came less than two weeks after launch, but the self-reported figure counts weekly users, not paying seats.


OpenAI’s agents hit 10 million users as ChatGPT Work push accelerates

OpenAI says the agentic side of ChatGPT has passed 10 million weekly users, a mark it reached less than two weeks after the debut of ChatGPT Work, its attempt to turn the chatbot into a tool that finishes a job rather than just describes one.

The figure, first reported by Bloomberg, spans the two products OpenAI now brands as agents: Codex, its coding assistant, and ChatGPT Work, the newer system that carries out multi-step tasks across a person’s apps and files.

What the number counts is worth pinning down, because the headline invites a generous reading. It refers to weekly active users of the two agentic products combined, not enterprise seats, paying customers, or the number of agents actually run.

It is also a company estimate rather than an audited metric, and OpenAI’s usage claims have drawn scepticism before, including over the self-reported figures behind its internal adoption research.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol' founder Boris, and some questionable AI art. It's free, every week, in your inbox. Sign up now!

Even allowing for that, the curve is steep. Codex passed 5 million weekly users at the end of May, reached 6 million on 12 July and roughly 8 million by the middle of the month, according to counts OpenAI executives posted as each threshold fell.

More than a million of those users now apply Codex to work outside software development, the company has said, and usage of the agentic tools as a whole rose about 2.5 times in the week after OpenAI shipped GPT-5.6, the model that runs ChatGPT Work. Sam Altman has also repeatedly reset usage caps as a nudge to push adoption higher.

ChatGPT Work launched on 9 July for Pro, Enterprise, and Edu subscribers, then reached Plus and Business plans within days.

Rather than answering a prompt and stopping, it splits a brief into smaller steps and works through them, generating spreadsheets, slide decks, documents, and small web apps while staying on a project for hours at a time.

The agent connects to services such as Slack and Google Drive, can schedule recurring tasks, and uses OpenAI’s Computer Use feature to operate desktop software on a user’s behalf.

Pricing follows the usage-based model already applied to Codex, and administrators on Enterprise and Edu plans can cap spending, restrict which tools it can touch, and review activity through a compliance interface.

For OpenAI, the milestone matters less as a measure of scale than as a signal of direction. The company has spent the year trying to move ChatGPT from a consumer habit into software that finance, sales, and operations teams pay to keep running, and it says nearly all of its own staff now use the agentic products daily.

Zapier, RingCentral, Virgin Atlantic, and Nvidia were named at launch as early customers, with finance teams using the tools to speed up month-end reporting and forecasting by automating the work of pulling data, reconciling figures, and building the presentations that follow.

The timing places OpenAI in a crowded contest. Google recently launched Gemini Spark, an always-on assistant that takes instructions through Gmail, while Anthropic has pushed its own coding and workplace agents.

All are chasing the same claim, that an AI system can be trusted to finish a task rather than merely draft it, and all are competing for the same enterprise budgets and the same finite pool of IT leaders willing to hand real work to software that still needs watching.

What the figure does not settle is whether the usage endures or converts into revenue at the scale OpenAI needs ahead of a widely reported public offering.

The company has been trialling other ways to make ChatGPT pay, including testing advertising, and a weekly-user count reveals little about how many of the 10 million sit on paid plans or how often they return once the novelty of a new model fades.

For now the number reads as a marker of momentum, self-reported and taken mid-race, rather than proof of a market OpenAI has already won.

Get the TNW newsletter

Get the most important tech news in your inbox each week.