OpenAI began rolling out the GPT-5.6 family, Sol, Terra, and Luna, to ChatGPT, ChatGPT Work, Codex, and the OpenAI API on July 9, ending a roughly two-week preview limited to partners the White House had cleared. The 24-hour global rollout arrives after the Department of Commerce’s Center for AI Standards and Innovation ran additional tests and OpenAI sent technical staff to Washington to answer reviewer questions, per Engadget’s reporting citing Axios.
The gating traces to an AI cybersecurity executive order the White House signed in early June, which asks companies to voluntarily submit their most capable models for government review 30 days before public release. GPT-5.6 is the first flagship system to move through that pipeline end-to-end. It’s also, functionally, the first stress test of whether “voluntary” pre-release review scales.
OpenAI isn’t pretending to like the arrangement. In its June 26 preview post, the company said the current process “keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them,” and added: “We don’t believe this kind of government access process should become the long-term default.” That’s polite frontier-lab dissent, filed with a straight face while the model sits in a Commerce Department review queue.
The specs read like a normal generational bump wearing regulatory clothing. All three models share a 1.05 million token context window and 128K max output. Sol is priced at $5 per million input and $30 output tokens; Terra at $2.50 and $15, pitched as half the price of GPT-5.5; Luna at $1 and $6. Prompt cache reads carry a 90% discount with a 30-minute minimum lifetime; cache writes bill at 1.25x the uncached input rate. The gpt-5.6 alias routes to Sol, and a Sol-on-Cerebras offering hits up to 750 tokens per second.
The benchmarks OpenAI chose to disclose tell you what the review was actually about. GPT-5.6 posts 53.5% on SecureBio’s Virology Capabilities Test, 60.0% on Molecular Biology, 68.4% on Human Pathogen Capabilities, and 68.3% on World-Class Bio, roughly nine points above GPT-5.5 across the suite. Sol also sets a state of the art on Terminal-Bench 2.1 and matches leading ExploitBench results using about a third of the output tokens of a rival system. Cheaper offensive-security reasoning is precisely the capability profile the executive order was written to notice.
The competitive backdrop tightened a day early. On July 8, SpaceXAI shipped Grok 4.5, which Elon Musk described as an “Opus-class model,” priced at $5 per million input and $25 output tokens. It’s not clear whether Grok 4.5 went through the same 30-day window. That asymmetry, if it holds, is the actual policy story: a review regime that binds the labs willing to be bound.
Sources
- https://openai.com/news/
- https://openai.com/index/previewing-gpt-5-6-sol/
- https://techcrunch.com/2026/06/26/openai-limits-gpt-5-6-rollout-after-government-request-says-restrictions-shouldnt-be-the-norm/
- https://techcrunch.com/2026/07/08/spacexai-releases-grok-4-5-which-elon-describes-as-an-opus-class-model/
- https://www.engadget.com/2210308/openai-rolls-out-gpt5-6-july-9/