OpenAI launches ChatGPT Work, an AI agent for repetitive tasks

OpenAI has launched ChatGPT Work, an AI agent designed to handle repetitive tasks on behalf of users. The company said it aims to expand ChatGPT's usefulness beyond simply answering questions, positioning it as a collaborative partner that works alongside users.

OpenAI announced Friday that it is releasing ChatGPT Work, a new work-focused AI agent.

ChatGPT Work is an AI agent built to take over users' repetitive tasks. Through a scheduling feature, users can instruct ChatGPT to carry out tasks automatically at set intervals or when specific events occur.

Users can also draw on information from applications outside ChatGPT to produce finished outputs such as spreadsheets, presentations, documents and web apps. The agent can break complex projects into smaller steps and execute them independently. Users can hand off an entire workflow to ChatGPT Work with a single request.

Starting Friday, ChatGPT Work is rolling out on web and mobile for Pro, Enterprise and Edu subscribers, with expansion to Plus and Business users expected within the next few days. The ChatGPT desktop app is also available from Friday to all users worldwide on Mac and Windows, across all subscription tiers including the free plan.

ChatGPT Work runs on GPT-5.6, OpenAI's latest frontier model, which officially launched Friday. According to OpenAI, Sol — one of three family models within GPT-5.6 — achieved top-tier performance in coding, knowledge work, cybersecurity and science. It also completes tasks using fewer tokens and at lower cost than OpenAI's previous models and competing offerings.

OpenAI said GPT-5.6 Sol scored 53.6 on "The Agent's Last Exam," a benchmark measuring the ability to perform extended professional tasks, outpacing Claude Fable 5 by 13.1 points. At a mid-level reasoning setting, it delivered 11.4 points higher performance at roughly one-quarter of the cost. On the Artificial Analysis Coding Agent index, Sol set a record high of 80, using half the output tokens and half the task time of Fable 5, with an estimated cost about one-third lower.

Sol also posted strong results on knowledge-work benchmarks. It set record scores of 92.2 percent on BrowseComp, which tests the ability to find hard-to-locate information on the web, and 62.6 percent on OSWorld 2.0, which evaluates performance on complex, long-horizon tasks in real computer environments. On OSWorld, it outperformed Opus 4.8 while using 85 percent fewer tokens.

Kim Kyung-hoon, head of OpenAI Korea, said ChatGPT Work shows that "AI is evolving beyond a tool that answers questions into a colleague that understands users' context and helps complete real work." He added that it would serve as "a first step in supporting Korean businesses and individuals to move beyond repetitive tasks, focus on what matters most, and turn good ideas into results."