OpenAI has unveiled GPT-5.6, but the new flagship model is only half the story. The bigger announcement is ChatGPT Work, a new experience that pushes ChatGPT beyond answering questions and into acting like a true AI coworker that can take on long-running projects across your apps, files, and workflows.
That distinction matters. AI companies have spent the past couple of years racing to post higher benchmark scores and bigger performance claims. OpenAI is still doing that with GPT-5.6, but this launch feels different. Rather than asking people to admire another smarter model, it is showing what that intelligence can actually do.
At the heart of the release is GPT-5.6, which comes in three versions. Sol is the flagship model built for the hardest work, Terra is positioned as the balanced option for everyday tasks, and Luna is the fastest and least expensive model in the family. OpenAI says all three deliver more useful work per dollar than GPT-5.5 by producing better results while consuming fewer tokens.
According to the company, GPT-5.6 Sol sets new state-of-the-art results across coding, scientific research, cybersecurity, computer use, and long-running knowledge work. OpenAI also claims it beats competing frontier models in several benchmarks while costing less to run, although, as always, benchmark charts only tell part of the story. Real-world use will ultimately determine whether those gains are as meaningful as they look on paper.
One of the biggest additions is a pair of new reasoning modes. Max gives GPT-5.6 more time to think through difficult problems before responding. Ultra goes much further by automatically coordinating four AI agents working in parallel. Rather than tackling a project step by step, the agents can split the workload, compare results, verify each other’s work, and produce a final answer more quickly. OpenAI says developers can build similar multi-agent workflows using new capabilities in its API.
Coding continues to be one of OpenAI’s strongest selling points. The company says GPT-5.6 Sol is its best programming model yet, outperforming GPT-5.5 while generating fewer output tokens and completing jobs faster. It also introduces Programmatic Tool Calling, which lets the model write and execute lightweight programs to coordinate tools, process intermediate data, and decide what information actually needs to be sent back to the model. That reduces unnecessary back-and-forth while lowering costs.
For many people, however, ChatGPT Work will likely be the headline feature.
Instead of responding to a single prompt and waiting for another, ChatGPT Work can stay focused on a project for hours if necessary. It can collect information from connected services including Slack, Microsoft Teams, Google Drive, SharePoint, Gmail, Outlook, Google Calendar, CRMs, and other business tools, then transform that information into finished presentations, spreadsheets, reports, marketing plans, dashboards, or other polished deliverables.
The idea is simple. Instead of asking ChatGPT dozens of questions throughout the day, you hand it an objective and let it work toward completing it.
OpenAI says you can ask ChatGPT Work to analyze a month’s worth of financial data, prepare for a sales meeting, review customer research, generate a marketing campaign, summarize conversations across multiple platforms, or create executive presentations. While it works, you can monitor its progress, answer questions, redirect its efforts, or approve important actions before it continues.
The company is also introducing Scheduled Tasks, which allow ChatGPT to repeat work automatically. For example, it can review new Slack messages every Monday morning, summarize overnight activity, refresh a recurring presentation, monitor customer feedback, or check websites for changes and generate reports without being asked each time.
Another interesting addition is Sites, now available in public beta. Rather than simply creating documents, ChatGPT can build interactive websites and lightweight web applications directly from a prompt. OpenAI says Sites can be used for dashboards, launch calendars, internal portals, project trackers, reports, and prototypes that remain editable and can even update as underlying information changes.
The desktop experience is changing as well.
OpenAI is merging the standalone Codex application into a redesigned ChatGPT desktop app for both Windows and macOS. The new application combines Chat, Work, and Codex under one roof while adding a built-in browser capable of gathering information from the web and interacting with online tools.
On desktop, ChatGPT also gains expanded Computer Use capabilities. That means the AI can interact with applications on your computer by clicking buttons, typing text, navigating websites, moving files, and carrying out repetitive tasks while you supervise the process. OpenAI says it can even use those capabilities as part of recurring Scheduled Tasks.
Developers are getting attention too. The updated desktop application supports multiple repositories inside a single project, faster computer use powered by GPT-5.6, inline editing for code reviews, and improved pull request workflows. Meanwhile, OpenAI plans to retire the standalone Atlas browser, folding what it learned from that project into ChatGPT itself, including a new Chrome sidebar extension that brings the AI directly into the browser.
OpenAI is also placing heavy emphasis on enterprise adoption. Administrators can decide which apps ChatGPT can access, what company data it can use, and which actions require approval. A new auto-review system examines sensitive operations before they are performed, while compliance tools allow organizations to monitor ChatGPT Work activity across their environments.
The company says GPT-5.6 is also its strongest cybersecurity model yet, helping security professionals identify vulnerabilities, validate patches, review code, analyze malware, and support defensive operations. OpenAI paired those expanded capabilities with what it describes as its most extensive safety effort so far, including approximately 700,000 GPU hours of automated red-team testing, extensive human evaluations, continuous monitoring, and stronger safeguards designed to reduce misuse without unnecessarily blocking legitimate security research.
Availability begins today, although not everyone will receive everything immediately.
GPT-5.6 is rolling out across ChatGPT, Codex, and the OpenAI API over the next 24 hours. ChatGPT Work launches first for Pro, Enterprise, and Edu subscribers on the web and mobile, with Plus and Business users following over the next few days. The redesigned ChatGPT desktop app is available globally today for Windows and Mac, with Chat, Work, and Codex included on every plan, including Free.
Developers can also access all three GPT-5.6 models through the OpenAI API. Pricing starts at $1 per million input tokens for Luna, $2.50 for Terra, and $5 for Sol, with corresponding output token pricing scaling from $6 to $30 per million tokens depending on the model.
GPT-5.6 undoubtedly raises the bar technically, but ChatGPT Work feels like the more important announcement. Better benchmarks are expected every few months. A ChatGPT that can independently organize information, monitor workflows, create polished deliverables, interact with software, and stick with a project until it’s finished represents a much bigger change.
Look, folks, if OpenAI delivers on those promises in day-to-day use, people may stop thinking of ChatGPT as something they ask questions and start thinking of it as something that actually helps them get work done.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz