OpenAI launches ChatGPT Work to automate office tasks

OpenAI unveiled ChatGPT Work on Thursday, an agent inside ChatGPT that can move across apps and files, stay on a project for hours and turn a goal into finished work. Built on GPT-5.6, the product is aimed at office users who want documents, spreadsheets, presentations and websites assembled with far less manual stitching between tools.

It can handle month-end budget variance analysis, turn source materials into a marketing brief or prepare a sales meeting. It can also continue through Scheduled Tasks when the user is away from a computer or phone. A new desktop application and a hosted-websites feature let users build and share sites directly through Work, while the rollout began on web and mobile immediately and was set to expand over the next few days.

The launch sharpened OpenAI’s contest with Anthropic as enterprise AI moves from chatbots that answer questions to agents that complete multi-step tasks. OpenAI has been pushing in this direction since April 22, when it introduced workspace agents for shared, long-running workflows in ChatGPT Business, Enterprise, Edu and Teachers plans. The July 9 release arrived alongside the GPT-5.6 family, which includes Sol as the flagship model, Terra as the balanced model and Luna as the lowest-cost model.

GPT-5.6 Sol scored 53.6 on Agents’ Last Exam, 13.1 points ahead of Claude Fable 5. The public release came after a limited preview requested by the United States government.

More than 5 million people use Codex every week and more than 1 million now use it for work outside software development. Zapier used ChatGPT Work to review thousands of leads each month and found seven figures in potential sales. In another internal workflow, one user could support roughly 50 product managers instead of one.