OpenAI ships ChatGPT Work and the GPT-5.6 family after 13-day Commerce review

The new agent runs multi-hour workflows across connected apps and files; the accompanying Sol, Terra, and Luna models are live in ChatGPT, Codex, and the API.

OpenAI on Wednesday launched ChatGPT Work, an in-chat agent designed to run multi-hour tasks across connected apps and files, and made the GPT-5.6 family (Sol, Terra, and Luna) generally available in ChatGPT, Codex, and the API. The launch arrives after a 13-day delay that Bloomberg attributes to the Trump administration asking for additional review before wider release.

That review is the more interesting fact. Per Axios, the Department of Commerce’s Center for AI Standards and Innovation ran further testing on GPT-5.6, and OpenAI sent technical staff to Washington to field questions. The White House said in a statement that no formal clearance was required and framed the process as voluntary. Read that carefully: a voluntary process that nonetheless held a product back for nearly two weeks is the working definition of soft-touch industrial policy, and it’s the kind of arrangement the pre-launch review of financial products under Dodd-Frank normalized after 2010. The optics of “voluntary” reviews rarely stay voluntary for long.

ChatGPT Work gathers context across a user’s connected apps and files, decomposes goals into steps, and returns finished documents, spreadsheets, slides, or web apps. It ships with Plan mode, configurable check-ins, action approvals, and Codex technology under the hood. Pro, Enterprise, and Edu customers get access first, with Plus and Business tiers following within days per Reuters.

The pricing tells its own story. Sol runs $5 per million input tokens and $30 per million output; Terra sits at $2.50 and $15; Luna at $1 and $6. Sam Altman claimed on CNBC that Sol is “54% more token efficient on agentic coding tasks,” and added that “every enterprise now is thinking about spend and the value they’re getting in exchange for AI.” Sol also introduces a max reasoning effort setting and an “ultra” mode that orchestrates subagents.

OpenAI says nearly 100% of its own teams now use ChatGPT Work and Codex, with month-end close and forecasting shrinking from days to hours. Angela Ferrante, head of enterprise marketing at Zapier, told Reuters her team’s lead-review system produced “seven figures in potential sales.”

The surrounding product cleanup is aggressive. Codex is merging into a unified ChatGPT desktop app for macOS and Windows, the standalone Atlas browser is being sunsetted, Sites launched in public beta to turn Work outputs into shareable web apps, GPT-5.4 retires July 23, and GPT-5.6 becomes the preferred model in Microsoft 365 Copilot. The consolidation looks less like a product roadmap than a company deciding which of its own experiments were narrative and which were infrastructure.

OpenAI ships ChatGPT Work and the GPT-5.6 family after 13-day Commerce review

Sources