OpenAI GPT-5.6 Sol, Terra, and Luna: July 9 Public Launch After First-Ever U.S. Government-Gated Approval

On July 9, 2026, OpenAI publicly launched the full GPT-5.6 model lineup — Sol, Terra, and Luna — after the U.S. Department of Commerce's Center for AI Standards and Innovation (CAISI) cleared the models for broad release, following additional testing ordered by the Trump administration ^RTA. The launch ended a two-week restricted preview that began June 26, 2026, when OpenAI released the models to only a small group of government-approved partners at the explicit request of the White House — the first time the U.S. government had preemptively gated an American AI company's model launch ^TOP.

Federal approval through CAISI and Trump's executive order

• CAISI clearance: The Commerce Department's CAISI conducted safety testing on the GPT-5.6 family and approved the broad rollout on July 7, 2026, clearing the way for the July 9 public launch ^TLA.
• Trump administration's role: The White House Office of Science and Technology Policy (OSTP) and the Office of the National Cyber Director (ONCD) jointly requested the staged rollout in June, citing national security concerns — making it the first documented case of the executive branch restricting a frontier AI model's commercial release ^AP. Axios reported the approval came after "additional testing" and meetings between OpenAI and government officials ^A.
• Prior restriction: From June 26 onward, OpenAI limited access to approximately 20 government-vetted organizations, accessible only through the API and Codex, with ChatGPT users entirely excluded ^QAY.
• Executive Order foundation: The Trump administration invoked Executive Order 14409, signed June 2, 2026, to require OpenAI to stagger the public release so the government could vet which partners get early access ^OL.

Sol's capabilities and pricing

• Capabilities: Sol is the flagship model, described by OpenAI as its "strongest model yet" ^MO. It scored 91.9% on TerminalBench 2.1, beating both Claude Fable 5 and GPT-5.5 ^L. On ExploitBench, Sol matched Anthropic's Mythos Preview using only one-third the output tokens ^L. All three GPT-5.6 models cleared "High" ratings in both biological and cybersecurity capability for the first time in any OpenAI model family — the core reason for government scrutiny ^AA.
• Pricing per 1M tokens ^AL:
  • Sol: $5 input / $30 output
  • Terra: $2.50 input / $15 output (half the cost of GPT-5.5 with competitive performance)
  • Luna: $1 input / $6 output (fastest, cheapest tier)
• New features: The lineup introduces a "maximum reasoning effort" setting and an "ultra mode" (also called "super mode") that uses sub-agents to parallelize complex tasks ^OL.
• Cerebras launch: GPT-5.6 Sol will also launch on Cerebras at up to 750 tokens per second in July, with initial customer access limited as capacity expands ^A.

Regulatory precedent

• First-ever government-gated AI launch: This was the first time OpenAI voluntarily restricted a model's commercial release at a direct White House directive, following the earlier precedent set when the Trump administration imposed similar restrictions on Anthropic's Mythos model ^OYT.
• OpenAI's position against permanence: OpenAI explicitly stated it does not want this kind of government access process to become the long-term default, writing: "We don't believe this kind of government access process should become the long-term default" ^LT.
• Emerging framework: The CAISI testing-and-approval pathway used for GPT-5.6 — federal safety review, conditional clearance, and staged rollout — mirrors the structure outlined in Trump-era executive orders on AI safety and is expected to serve as a template for future frontier model releases ^TA.

Key caveat

Reuters noted on July 8 that Axios's report on the Commerce Department approval could not be independently verified by Reuters at that time ^RL. However, the July 9 public launch has since been confirmed across multiple outlets including OpenAI's own announcements and major news organizations ^RTA.