OpenAI ships ChatGPT Work, the agent that returns finished files
The new agent runs for hours across your apps and hands back spreadsheets, decks and working sites. Codex folds into one desktop app, and every tier gets it, including Free.
- OpenAI launched ChatGPT Work on July 9, an agent that takes a goal, works across a user's connected apps and files for hours, and returns finished deliverables: spreadsheets, slides, documents, reports, and working web apps.
- It runs on GPT-5.6, released to general availability the same day in three variants: Sol, Terra and Luna.
- The Codex app merges into a new ChatGPT desktop app on every plan, including Free; the existing app is renamed ChatGPT Classic.
- Rollout starts with Pro, Enterprise and Edu on web and mobile, with Plus and Business following over the next few days.
OpenAI spent July 9 repositioning ChatGPT from a place where you talk about work to a place where the work comes back done. ChatGPT Work, announced alongside the general availability of GPT-5.6, is an agent that accepts a goal, runs across a user's connected apps and files, and returns the finished artifact rather than a reply.
From chat to deliverables
The deliverables framing is the story. Per OpenAI's own documentation, Work sessions can run for hours and produce spreadsheets, presentations, documents, reports, and shareable web apps. That is the same product promise Anthropic has been making with Claude Cowork, and the same one Salesforce sells into support organizations with Agentforce. The difference is distribution: ChatGPT Work lands inside the largest consumer AI product in the world, on every tier.
Sam Altman leaned on the economics in his launch remarks, telling Forbes that GPT-5.6 is 54 percent more token efficient on agentic coding, a figure OpenAI has not yet published in its own materials. "Every enterprise now is thinking about spend and the value they're getting in exchange for AI," he said.
The Codex consolidation
Alongside Work, OpenAI is folding the Codex app into a new ChatGPT desktop app that puts Chat, Work and Codex in one place on every plan, including Free. The current desktop app becomes ChatGPT Classic. Consolidation is the tell: OpenAI wants the agent, the chat and the coding surface to be one product with one habit loop, not three apps competing for the dock.
The pause before the launch
The GA itself arrived later than planned. OpenAI staggered the GPT-5.6 rollout after a request from the Trump administration, and Altman described making "many changes" after a "collaborative back and forth" with the government. The models are now available across ChatGPT and the API; how the safety community reads that timeline is its own story.
For the agent economy, the takeaway is simpler. The category leader just put a deliverables agent in front of every free user on earth, priced the model underneath it on a three-step ladder, and made the whole thing the default window. Agents stopped being a feature yesterday.