Enterprise

OpenAI ships GPT-5.6 Sol, Terra, and Luna after 12-day White House gate, unveils ChatGPT Work

The frontier model family goes live across ChatGPT, Codex, and the API on July 9, alongside an agentic assistant plugged into Slack, Teams, Google Drive, and CRMs.

By Asher Glick · Enterprise software desk · 2026-07-09 · 4 min read

Photo: Unsplash / David Everett Strickler — The White House in Washington, DC

OpenAI made GPT-5.6 broadly available on July 9 across ChatGPT, Codex, and the API, twelve days after federal reviewers restricted the model’s preview to roughly 20 government-vetted organizations. The frontier family ships in three tiers (Sol, Terra, Luna), and it arrives paired with ChatGPT Work, an agentic enterprise assistant that plugs into Slack, Microsoft Teams, Google Drive, and SharePoint.

The 12-day gate is the story underneath the launch. In early June, a Trump executive order asked frontier labs to voluntarily submit their most powerful models for federal review up to 30 days before release. On June 26, the Office of the National Cyber Director and the Office of Science and Technology Policy invoked that framework, citing Sol’s cybersecurity capabilities as reason to keep access narrow. Axios first reported the delay; Reuters confirmed it. The Department of Commerce ultimately signed off on the broad rollout, per those same reports. This is the first real live-fire test of the voluntary regime, and it worked less as a brake than as a staging area.

Pricing signals where OpenAI wants each tier to live. Sol runs $5 in / $30 out per million tokens and is built for long-horizon agentic work; on Cerebras infrastructure it’ll hit up to 750 tokens per second in July, which matters because extended agent loops become economically viable at that speed. Terra sits at $2.50 / $15 for everyday use. Luna, at $1 / $6, is the volume play.

ChatGPT Work is the deployment surface for all of it. OpenAI says the assistant runs multi-step projects for hours without continuous supervision, and that nearly 100% of internal teams (finance and sales included) already use it alongside Codex. One cited sales workflow compresses discovery-to-proof-of-concept from weeks to 24 hours. Enterprise and Edu admins gate access, connected tools, and permitted actions; a Compliance API surfaces conversations for review; and OpenAI reports that an auto-review layer blocked 100% of protected-data extraction attempts during adversarial red teaming.

The shape here echoes the post-Dodd-Frank stress-test regime of 2011, where regulatory pre-clearance became a marketing asset rather than a constraint. A 12-day pause, ending in federal approval, isn’t friction. It’s a certification stamp, and OpenAI now has one competitors don’t.

Sources