OpenAI Group PBC today launched a new “agentic” tool called ChatGPT Work as it announced the global rollout of its most advanced model family so far in GPT-5.6.

ChatGPT Work is a new mode within ChatGPT that’s designed to perform actions autonomously across user’s connected applications, files, web tools, desktops and recurring workflows. It’s meant to expand ChatGPT’s capabilities, so it goes beyond simply answering prompts to executing more complex tasks. For instance, it can be used to create new spreadsheets, slide presentations, documents, dashboards, web applications and other content from a single prompt.

In a blog post, OpenAI said ChatGPT Work is initially being made available to Pro, Enterprise and Edu subscribers through its web and mobile applications. Pro and Business tier users will get access in the “coming days.” Meanwhile, the updated ChatGPT desktop app has been made globally available on Windows and Mac computers, integrating the basic Chat and Codex experiences with Work into a single service that’s accessible across all plans, including for free users. As part of the update, the Codex app will now live within the ChatGPT desktop app.

ChatGPT Work is powered by GPT-5.6, which is OpenAI’s latest frontier model, also announced today. It incorporates the complete Codex experience to enable multistep execution of any tasks with a coding element. According to OpenAI, the new work-based app has the ability to decompose larger goals and break them down into smaller tasks and work on them over extended timeframes, seeing guidance from human users when necessary. Before it carries out “sensitive actions,” it will ask a human for approval.

One of the most impressive things about ChatGPT Work is its available armory, with plugins for tools such as Slack, Microsoft Teams, Gmail, Google Drive, Salesforce, Sharepoint, calendars, project tracking and management tools and other work-related platforms. Users can invoke any specific applications they want the agent to use simply by typing “@” and the name of its plugin.

Alongside ChatGPT Work, the company also debuted a new capability called Sites in public beta for Plus, Pro, Business and Enterprise subscribers. This makes it possible to transform work materials into shareable, interactive websites or web apps, and encompasses internal portals, dashboards, calendars, project trackers, prototypes and interactive reports. ChatGPT can test these sites and update them automatically whenever the source information changes.

In another update, OpenAI said the ChatGPT application now gets its very own integrated browser with Computer Use functionality. This makes it easier for ChatGPT to get context from the internet, access and use browser-based tools and navigate across Google Workspace and Microsoft 365. Meanwhile, the ChatGPT Chrome extension is being enhanced, so that ChatGPT is now positioned in the sidebar.

OpenAI said ChatGPT Work is aimed at knowledge workers, enterprises, students, developers and business executives that need their AI tools to do more than just respond to isolated queries. In its lengthy blog post, it offered examples of how the new tool might be put to use in areas such as sales, marketing, finance, data analytics, engineering and business operations. Workers at companies such as Nvidia Corp., Zapier Inc., RingCentral Inc. and Virgin Atlantic Airways Ltd. have been using ChatGPT Work for some time in its beta testing phase. Use cases include reviewing thousands of sales leads, checking new products for launch readiness across Jira and go-to-market plans, comparing airline passenger experiences and automating event preparation for Nvidia GTC 2026.

VIDEO<span></span>

Organizations who have reason to be worried about unleashing AI agents across their systems can use ChatGPT Enterprise controls to manage ChatGPT Work’s access to browser use, plugins, connected tools and more. They can preauthorize what sensitive actions it’s allowed to take, implement usage limits and so on, OpenAI said.

GPT-5.6 now broadly available

Chief Executive Sam Altman said the GPT-5.6 model family is being made available in three variants, with the flagship offering Sol designed for advanced reasoning and coding tasks. The other two, Terra and Luna, are targeted at enterprise workloads and high-volume, low-cost applications that need less processing punch.

Altman told CNBC in an interview that GPT-5.6 shows a 54% improvement in token efficiency compared to its predecessor GPT-5.5 when carrying out agentic coding tasks. That could significantly reduce the costs of using ChatGPT for automated software development. Agentic coding has emerged as one of the most critical enterprise use cases for AI tools, enabling anyone to create software using only natural language prompts.

The 54% improvement in token efficiency means that GPT-5.6 will use fewer tokens than older models. Tokens are the units used by AI systems to process and generate text, and when a model uses lower amounts of them, it means reduced computational costs. “Every enterprise now is thinking about spend and the value they’re getting in exchange for AI, and this is what we really want to do,” Altman said.

The launch of GPT-5.6 comes at a pivotal moment for OpenAI, which is facing intensifying competition from rivals such as Anthropic PBC, SpaceXAI Corp. and Google LLC. Elon Musk’s company released its own new model, Grok 4.5, just a day earlier, and that also delivered a big improvement in token efficiency in order to encourage further enterprise adoption.

GPT-5.6’s launch was notable because the model was heavily scrutinized by the U.S. government before being made available to the public. The review process was overseen by officials such as the Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick and U.S. National Cyber Director Sean Cairncross. According to Altman, GPT-5.6 was reviewed in a “collaborative” process, which involved the company working with federal agents to address a number of questions regarding its safety.

“If you want broad access, which we do, and you have powerful models, you really want to be confident in your safety claims,” Altman said. “Because otherwise the world is going to get uncomfortable very fast.”

Image: OpenAI