OpenAI has launched GPT-5.6, the newest generation of its frontier models, and unveiled ChatGPT Work, a workspace product built for teams, in a release that is as notable for how it happened as for what shipped. GPT-5.6 comes in three tiers, Luna, Terra, and Sol running from least to most capable, plus a top end Sol Ultra, and it is trained on data through approximately May 2026. On the usual measures it is a strong step up, but the numbers are not the headline this time.
Start with the lineup, because it tells you how OpenAI is thinking about cost and capability. Terra is positioned as competitive with the previous GPT-5.5 while costing about half as much, which pushes the price of a capable model down, and Sol and Sol Ultra sit at the top, with Sol Ultra clearing 90 percent on the TerminalBench agentic coding benchmark. The practical read is a family that lets developers dial in the tradeoff they want, a cheaper workhorse in Terra and a maximum capability option in Sol, rather than a single one size fits all model.
The most unusual part of the launch is how it was released. OpenAI says that as part of ongoing engagement with the US government it previewed the plans and capabilities of the models ahead of launch, and at the government's request it started with a limited preview for a small group of trusted partners, whose participation was shared with the government, before releasing more broadly. The trigger for that caution is cybersecurity. OpenAI describes GPT-5.6 Sol as its most capable model yet for cyber tasks, one that shifts the frontier on long horizon security work including vulnerability research and exploitation. That kind of capability is dual use, it helps defenders find and fix flaws and it helps attackers find and exploit them, and the staged, government visible rollout is OpenAI's answer to that tension.
The second announcement is ChatGPT Work, which OpenAI frames as its biggest work focused update to date. Rather than a smarter chatbot, it is pitched as a dedicated workspace where a team collaborates with the AI on projects, kept separate from personal chats and stored files, with workspace controls, stronger context handling, and the kind of multi step project execution that Codex has shown, all running on the new GPT-5.6 models. It is OpenAI leaning harder into the enterprise, where the contest to become the place people actually get work done, not just the tool they occasionally ask questions, is heating up fast. During the preview period the models themselves are reaching developers through the OpenAI API and through Codex.
What makes the day matter is the pair of firsts underneath the version number. On the capability side, a leading lab openly calling its newest model its strongest for security relevant tasks, and choosing to gate the release with government visibility rather than simply flipping it on, is a real shift in how frontier models get shipped, with safety and security review acting as a release gate instead of a footnote. On the product side, ChatGPT Work is a signal that the prize OpenAI is chasing next is the workplace itself. The benchmark bragging is present as it always is, but the story here is the launch process and the push into where teams work, and both are more consequential than another point on a chart.