OpenAI is making GPT-5.6 available beyond a restricted preview after receiving approval from the Trump administration, The Verge reported. The shift ends a roughly two-week period in which access to the model was limited to organizations cleared by the government.
The public release lands alongside ChatGPT Work, a new agent product OpenAI says combines ChatGPT with capabilities from Codex. The pitch is aimed less at engineers staring down a repo and more at office workers who want the model to assemble work products from the digital mess they already use every day.
According to OpenAI, ChatGPT Work can pull context from apps, files, and workflows selected by the user, then produce documents, spreadsheets, presentations, and web apps. OpenAI also said a unified plugin directory lets ChatGPT connect with services including Slack, Gmail, Google Drive, calendars, and customer relationship management tools.
That mechanism is the product: the model is not just answering in a chat box. OpenAI is trying to give it controlled access to workplace inputs and output formats, then have it act across them. The company’s claims still need real-world testing, because agent demos have a habit of looking much cleaner in launch materials than they do inside actual corporate permission systems, half-maintained spreadsheets, and Slack threads with 19 contradictory instructions.
Who gets it first
OpenAI said Mac and Windows users worldwide, including people on the free ChatGPT tier, should be able to use ChatGPT Work and GPT-5.6 through the ChatGPT desktop app right away.
Access is more staggered on phones and the web. OpenAI said Pro, Enterprise, and Edu users will receive access first. Plus and Business users are scheduled to get it over the next few days. The company said the rollout is global, gradual, and expected to reach full availability over the next 24 hours.
GPT-5.6 is described by OpenAI as a model suite made up of Sol, Terra, and Luna. ChatGPT Work runs on that suite. OpenAI is putting the most weight on Sol, which it claims is the strongest of the three and sets a higher bar for intelligence and efficiency, including in coding, cybersecurity, science, and computer-use tasks.
The company is also presenting GPT-5.6 as a cheaper option than rival top-tier models. That claim arrives as AI companies face pressure over the cost of running advanced models and the prices passed on to customers, The Verge reported.
The agent race keeps getting noisier
OpenAI is not alone in trying to make an AI agent useful to ordinary users rather than just impressive in a lab clip. The Verge reported that OpenAI, Anthropic, Google, and Apple have all been pushing toward broader agent products, with mixed results. The open-source agent OpenClaw also helped raise expectations for what these tools should do.
ChatGPT Work directly competes with Anthropic’s Claude Cowork, which The Verge described as combining Claude with Claude Code. OpenAI’s bet is that binding ChatGPT, Codex-style execution, and GPT-5.6 into one product will make the agent feel less like a chatbot with a clipboard and more like software that can finish a task. OpenAI has made the claim. Users now get to find the seams.
This story draws on original reporting from The Verge.
