OpenAI on July 9, 2026 publicly released GPT-5.6, its most powerful model family yet, alongside ChatGPT Work — an AI agent designed to carry out whole jobs rather than just answer questions, according to Reuters and Axios.
GPT-5.6 comes in three tiers: Sol, the flagship; Terra, a mid-range option for everyday work; and Luna, a fast, low-cost version. Axios reports Sol also includes a new "ultra" mode. OpenAI listed pricing at $5 per million input tokens and $30 per million output for Sol, $2.50 and $15 for Terra, and $1 and $6 for Luna.
CEO Sam Altman said the model is 54% more token-efficient relative to rivals, per CNBC, and told Livemint the changes were aimed "entirely" at cost and speed, calling it "also much faster."
ChatGPT Work merges ChatGPT with OpenAI's Codex coding tool and runs on GPT-5.6. The Decoder reports it can independently handle complex projects across apps like Google Drive, Slack, and Salesforce, and is available on web, mobile, and desktop. Bloomberg says the agent can field tasks for hours. Livemint frames the launch as a bid to challenge Microsoft and Google.
The rollout followed U.S. government review. The Verge reports GPT-5.6 was earlier confined to a "limited preview" for government-approved organizations before the Trump administration cleared a public release; Altman said OpenAI "made many changes." TechCrunch noted it remains unclear exactly how the government decided the model was safe.
Not everyone is impressed with the packaging: M.G. Siegler of Spyglass called the new ChatGPT "Super App" a "tangle of toggles and strange UI decisions," saying the Mac app is "a mess."
Why it matters: OpenAI is pushing from chatbots that talk toward agents that act, a shift with real stakes for how everyday office work — and the jobs behind it — gets done.