The news
On July 9, 2026, OpenAI launched ChatGPT Work, an agent inside ChatGPT that takes action across a user's apps and files and can stay on a project for hours. It runs on GPT-5.6, a new model family OpenAI released the same day.
OpenAI said the agent breaks a goal into smaller steps and completes them on its own, producing finished spreadsheets, slide decks, documents and web apps, as reported by PYMNTS and TestingCatalog. Yahoo Tech described it as combining Codex, OpenAI's coding agent, with ChatGPT. According to TestingCatalog, it connects to tools including Slack, Microsoft Teams, Gmail, Google Drive, SharePoint and Salesforce, and on desktop it can use a built-in browser and operate the computer directly.
On the web and mobile apps, OpenAI began rolling out ChatGPT Work to its Pro, Enterprise and Edu plans on July 9, with Plus and Business plans to follow over the next few days, PYMNTS reported. In the ChatGPT desktop app it was available on every plan, which TestingCatalog said includes the Free tier.
For Enterprise and Edu customers, TestingCatalog reported that administrators can control plugin access, browser use, connected tools, network access, sensitive actions, spending and usage limits. It said an auto-review step checks critical actions before they run.
GPT-5.6 comes in three tiers: Sol, which OpenAI positions as its strongest model; Terra; and Luna, the lowest-cost option. Chief executive Sam Altman said Sol is 54% more token efficient on agentic coding jobs, according to PYMNTS. Yahoo Tech reported that the White House had asked OpenAI to delay the release, and that OpenAI first limited the models to trusted partners before the public launch. Yahoo Tech also reported that OpenAI is shutting down its Atlas browser and folding what it learned into ChatGPT Work.
The numbers
- GPT-5.6 model tiers
- 3 (Sol, Terra, Luna)
- Sol token efficiency on agentic coding jobs
- 54% better (Sam Altman, per PYMNTS)
- First plans on web and mobile
- Pro, Enterprise, Edu
- How long one task can run
- Hours, per OpenAI
Why CEOs should care
An agent that returns a finished deck or financial model changes the unit of value from a chat answer to a completed deliverable. Operators should test ChatGPT Work on tasks their teams already do well, time it against current effort, and check the output line by line. That comparison, not the demo, shows which contractor, agency or analyst hours it could replace. Altman told CNBC that enterprises are weighing AI spending against the value they get, as reported by PYMNTS; CFOs should ask how long-running agent tasks count against plan limits and whether usage is billed separately.
For CISOs and IT leaders, the risk sits in the connectors. An agent that can read Gmail, SharePoint and Salesforce and then act on what it finds has more reach than any single employee tool. Before switching it on, decide which connectors are allowed, which actions need human approval, whether the agent may reach the open internet, and how its activity is logged. Desktop access on every plan, including Free, means staff could run the agent on work files outside company controls, so check endpoint and acceptable-use policies now.
Boards should ask who is accountable when an agent-built report or spreadsheet contains an error, and whether review steps keep up with output that arrives faster than people can check it.
The bigger picture
The launch moves ChatGPT from answering questions toward carrying out whole assignments. It also showed that frontier releases now face government attention before they ship: by Yahoo Tech's account, the White House asked for a delay, and OpenAI said it spent multiple weeks pressure-testing and hardening its system before the public launch.
For software vendors, an agent that builds its own spreadsheets, dashboards and web apps from company data could compete with tools sold per seat, since the same work may be done inside a ChatGPT plan the customer already pays for.
What happened next
On September 10, 2026, OpenAI added a Data plugin that lets Work and Codex users analyze connected business data and build interactive dashboards, according to OpenAI's ChatGPT release notes as compiled by Releasebot. On September 22, it made GPT-6 Sol and GPT-6 Luna available in ChatGPT Work and Codex, separate from the models used in regular chat, and on September 23 voice in Work gained plugin support on web, iOS and Android.
What to watch: how OpenAI meters long-running Work tasks by plan, and whether it adds more audit and approval controls as enterprises connect the agent to core systems.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made with our newsroom’s technology tools, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is not published automatically; it is held for the editor, who decides whether it is fixed, published or dropped.
- Archive story
- It was written after the event. Its dates were checked against the date of the event, including the dates of any later developments it reports.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
