OpenAI head of core products Thibault Sottiaux says ChatGPT Work has reached 20 million users, and that the company is deliberately bundling agentic capability into the $20-a-month ChatGPT Plus tier rather than pricing it as a separate enterprise product. Sottiaux, who runs API, agent infrastructure, enterprise, all of ChatGPT and Codex, framed the strategy in an interview this week as a diffusion play: take a tool built for developers and repackage it for every white-collar worker, at consumer pricing. The number to watch is the gap between what a Plus subscriber pays and the compute those subscribers now consume through agents.
OpenAI has responded to that gap with Luna, an 80% price cut that Sottiaux called a permanent correction on the underlying token economics. He said the goal is for a user to wake up six months from now and be able to do the same work with less spend, and to keep folding new utility into the same $20 price point rather than raising prices with each capability release. That is a different pricing philosophy from most of the frontier — competitors have generally lifted headline prices as models improved.
The product itself is a repackaging of Codex, OpenAI's software engineering agent, into a workplace surface that runs on mobile, web and voice.
“We wanted to bring the power of coding agents to everyone, and so this is an exercise in taking something that was made for technical people, and then packaging it in a way that is like safe and delightful to use”— Thibault Sottiaux, OpenAI head of core products
Key facts
- 01ChatGPT Work has reached 20 million users, according to head of core products Thibault Sottiaux.
- 02OpenAI cut API prices 80% with Luna, which Sottiaux describes as a permanent price correction.
- 03ChatGPT Work is bundled into the $20-a-month Plus plan rather than sold as a separate enterprise tier.
- 04Sottiaux runs API, agent infrastructure, enterprise, all of ChatGPT and Codex, reporting to Greg Brockman.
- 05Sottiaux says users should be able to do the same work with less spend six months from now.
Sottiaux said the design philosophy is to get out of the way of the model. Rather than putting A/B choices in front of the user, ChatGPT Work is engineered to let the model take the lead — draft the slides, run the deep research, process the document dump, and hand back a finished artifact. Wharton professor Ethan Mollick has drawn the contrast with Anthropic's Claude Cowork, which pushes more decision points onto the human. Sottiaux's framing is that adoption is settling the argument.
ChatGPT Voice is part of the same bet. Sottiaux said voice usage has grown quickly because it removes the last piece of interface friction — users can just talk to the agent the way they would talk to a colleague, and the model adapts to the human rather than the reverse. The bet is that this is what pulls the non-technical majority in.
Adoption is the argument OpenAI is making to skeptics.
The 20 million figure covers ChatGPT Work specifically, and Sottiaux pointed to it as evidence that mainstream knowledge workers are willing to hand off multi-step tasks to an autonomous agent. He described the product roadmap as a process of discovery — OpenAI pushes model capability, then finds out what the model is good at, then builds surfaces around those strengths. GPT 5.6, he said, was a step up in professional work: long-document processing, quality slide generation, deep research, the kind of output a consultant or analyst would produce.
The pricing question is the one the industry is actually watching. A Plus subscriber running Codex-style agents through ChatGPT Work is consuming far more tokens than a chat user typing prompts, and the delta between the $20 monthly price and the underlying compute cost is what makes analysts nervous.
“This is something that will continue. Our goal is to, over time, include more utility in the same dollar amount.”— Thibault Sottiaux, OpenAI head of core products
Sottiaux's answer is that efficiency gains outpace usage growth. Luna's 80% cut is the first proof point; he expects more to follow as OpenAI keeps pushing on the frontier of inference efficiency. In effect, he is arguing that the unit economics of consumer AI improve fast enough to absorb heavier agentic workloads without breaking the $20 tier — a claim CFOs at OpenAI and its competitors are all making in slightly different forms.
The other pressure point is trust. ChatGPT Work rolled out iMessages access this week, on top of email integration, and Sottiaux acknowledged the obvious question about handing an agent the keys to a user's inbox. His response leaned on OpenAI's safety stack and published benchmarks rather than on new product-level guardrails, which will not fully answer enterprise buyers weighing agent deployments against data-handling policies. It is the counterweight OpenAI will keep getting asked about as ChatGPT Work moves deeper into work surfaces.
The strategic read: OpenAI is trying to own the application layer for knowledge work at consumer pricing before Anthropic's Claude Cowork or a Microsoft Copilot bundle can build the same distribution. Twenty million users on ChatGPT Work is a strong opening number, and pairing that with an 80% price cut on the underlying tokens is a signal that OpenAI intends to compete on cost as well as capability. The risk is that the same bundling strategy that wins the mainstream also caps how much revenue any individual user can generate — and that at some point the CFO question Sottiaux was asked about the token gap becomes OpenAI's question too.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
