OpenAI released its GPT-5.6 model family and a new autonomous workplace agent called ChatGPT Work on Thursday, ending a rollout delay imposed last month by the Trump administration over the flagship model’s cybersecurity capabilities. The dual launch reframes what had briefly looked like a regulatory setback as a demonstration that federal review and enterprise ambition can now share a schedule.

The family arrives in three tiers. Sol, the flagship, runs at $5 per million input tokens and $30 per million output tokens. Terra, positioned for everyday work, is priced at $2.50 and $15. Luna, the budget variant, comes in at $1 and $6. OpenAI says Sol scores 92.2 percent on BrowseComp and 62.6 percent on OSWorld 2.0, surpassing Anthropic’s Opus 4.8 on both, and does it while emitting 85 percent fewer output tokens on the latter. Sam Altman told CNBC that Sol is 54 percent more token-efficient on coding tasks than its predecessor.

The federal detour is the more revealing part of the story. Engadget, citing Axios, reported that the delay lifted after additional testing by the Department of Commerce’s Center for AI Standards and Innovation, an agency now functioning as a de facto gatekeeper for frontier releases. Anthropic was reportedly ordered to block foreign-national access to its Mythos and Fable cybersecurity models around the same window. Altman told Bloomberg that OpenAI made “many changes” during those talks, a phrase whose vagueness is itself the point.

OpenAI says it ran its most intensive pre-release safety evaluations to date, roughly 700,000 A100e GPU hours of automated red teaming, and that Sol’s safeguards block about ten times more potentially harmful activity than earlier models.

ChatGPT Work, rolling out Thursday on web and mobile with integrations across Slack, Notion, Microsoft 365 and Google Drive, is a direct answer to Anthropic’s Claude Cowork, which launched in January. Reuters framed it that way explicitly. Bloomberg notes enterprise sales are simply more lucrative than consumer products, and TechCrunch has long described Anthropic as the industry’s enterprise-focused underdog, a position OpenAI is no longer willing to concede. Microsoft, which has designated GPT-5.6 the preferred model inside 365 Copilot, gets the distribution surface it wanted.

The pattern is legible now. Frontier labs ship on Washington’s clock, price aggressively against each other, and race to embed inside the software people already open at 9 a.m.

Sources