ChatGPT Merges Chat and Work, Following Claude’s Playbook
Set Trending Topics as a preferred source on Google.
Four months after launching ChatGPT Work, OpenAI is scrapping the Work tab again, following Anthropic’s playbook for the second time this year.
In an email to subscribers, OpenAI announced that in the coming weeks it will test a “unified experience” that brings the Chat and Work tabs in ChatGPT together, starting with a small group of users. The idea: Think through an idea and then have it carried out right away, “all in one place.” Users will be able to bring in their files and plugins and ask ChatGPT to build a website, create a presentation or complete tasks on the web, without first choosing between chatbot and agent. Codex, the coding tool, remains a separate area in the desktop app. OpenAI wants to use the test group’s feedback before rolling the change out to everyone.
One Allowance for ChatGPT and Codex
The second part of the email is likely to spark more debate. As part of the overhaul, usage of ChatGPT and Codex will count toward a single shared allowance. Until now, Chat, Work and Codex ran on separate counters, with additional limits for images, voice and file uploads. OpenAI stresses that it is preparing the transition “carefully” and expects most users will not have to wait for their allowance to reset more often than they do today. Consumption depends on the model and the task: More advanced models and complex agentic tasks cost more. Unlimited use of Instant, the fast default model, remains included in the plan and does not count toward the allowance. Under Settings and Usage, users will be able to see how much of their total allowance is left and when it resets. Those who run out can wait, upgrade or buy credits.
The move does not come entirely out of the blue. Tibo Sottiaux, who runs ChatGPT and Codex at OpenAI, said on Lenny Rachitsky’s podcast in early October that the company planned to merge the Chat and Work surfaces. Shortly afterward, some subscribers saw a notice at checkout saying that one limit would apply to ChatGPT and Codex starting in December. On Reddit, threads on the topic gathered hundreds of votes within hours, with most comments asking how exactly usage would be counted, and some Plus users saying they would cancel. OpenAI has a reputation to lose on limits: When ChatGPT Work launched in July, the Work and Codex allowances were reset twice in a single day, and in August the company added up to 50 percent more usage. As a buffer ahead of the merger, OpenAI recently doubled the Work and Codex allowances for its Pro tiers (Pro 100, Pro 200 and Pro 500, at $100, $200 and $500 per month, roughly €86, €172 and €430).
Anthropic Has Already Been Through This
Anyone who knows the pattern will recognize it immediately. When OpenAI launched ChatGPT Work in July, it was a copy of Claude Cowork, which Anthropic had introduced earlier in the year, almost feature for feature: agent instead of chatbot, results instead of answers, connected apps instead of isolated conversation. And the next step has already been mapped out, too. In mid-September, Anthropic merged the Chat and Cowork tabs in Claude. Since then, users no longer pick a mode; Claude decides on its own whether a question needs a quick answer or a multi-step job involving research, code and files. Anthropic justified the move by saying customers often did not know which tab to choose. At the same time, Docs and Slides arrived in beta, while Claude Code stayed a separate area, just as Codex now does at OpenAI.
Anthropic was earlier on the allowance as well. With Claude, according to the support documentation, all activity counts toward the plan’s usage limits, and longer agentic tasks with web search, code and files consume more than a quick question. The fact that Claude Code has used the same five-hour and weekly limits as chat since it was included in the Pro and Max plans has drawn criticism of Anthropic too, with heavy users regularly complaining that a coding session drains their chat allowance. Anthropic responded in the spring with doubled five-hour limits and temporarily higher weekly limits. It is exactly this tension, one pool for everything and the question of how expensive an agentic task is within it, that is now catching up with OpenAI.
Why Both Are Doing the Same Thing
Behind the lockstep lies the same calculation. Agents that work for hours, build websites and click through browsers consume many times the compute of a chat; for Claude Cowork, industry estimates put the figure at five to twenty times. Separate pools for chat, agent and coding are hard to price cleanly, while a shared allowance shifts the decision about what the compute is spent on to users. For OpenAI, there is the added fact that Codex, by its own account, is used by more than five million people a week and has become a cost factor. At the same time, the mode switch, where users have to decide in advance whether they need a chatbot or an agent, is a product problem both companies would rather leave to the model.
The counterargument comes from users themselves. Anyone who previously used a $20 (about €17) Plus subscription for chat and Codex on the side effectively had two allowances; soon there will be one. OpenAI’s assurance that “most” users will not have to wait more often leaves open what happens to the rest, and the reference to purchasable credits hints at where things are heading for heavy users. That Instant remains unlimited softens the blow for casual users, but not for those who subscribed to ChatGPT Work precisely because it works for hours with the most powerful models. Whether OpenAI fares better than Anthropic, whose limit debates have dragged on for months, will become clear once the small test group grows larger.
