GPT-5.6 ships to everyone after a 13-day government gate, and Grok 4.5 lands 24 hours ahead of it

OpenAI moved GPT-5.6 to general availability across ChatGPT, Codex, and the self-serve API on July 9, ending a 13-day preview window that the company said had been restricted to a narrow set of partners at the request of the U.S. government. The three tiers, Sol, Terra, and Luna, went live simultaneously. Twenty-four hours earlier, SpaceXAI shipped Grok 4.5, its first release since going public and folding in Cursor.

The government gate is the story most worth pausing on. Frontier releases have long been shaped by informal coordination with Washington on cyber-capability grounds, but June 26 through July 9 marks the first time a sitting administration has publicly held a US model back and then let it out on a scheduled date. CNBC reported the Trump administration sought the limited rollout for cyber-capability reasons. Anthropic went through a similar episode late last month with Claude Fable 5 and Mythos 5, since lifted. The precedent isn't quiet anymore; it's calendared.

OpenAI's launch post frames the previous restriction as a cost. Keeping the preview closed, the company wrote, "keeps the best tools from users, developers, enterprises, cyber defenders, and global partners." That's narrative management doing real work, a company that spent two weeks accepting a government hold now writing the case, in its own voice, for why holds are bad.

The pricing, per TechCrunch's read of OpenAI's materials: Sol at $5 input and $30 output per million tokens; Terra at $2.50 and $15; Luna at $1 and $6. Grok 4.5 launched at $2 and $6, which puts it in a straight line against Luna.

The benchmark theater is aimed squarely at Anthropic. OpenAI claims Sol scores 80 on the Artificial Analysis Coding Agent Index, 2.8 points above Fable 5, while using less than half the output tokens, less than half the time, and running about a third cheaper. Terra sits just above Fable 5 on the same index. Luna, OpenAI says, outperforms Opus 4.8. Sam Altman told CNBC that Sol is 54% more token efficient than its predecessor on agentic coding tasks, which is the efficiency claim doing the real economic work, inference margins, not leaderboard points.

OpenAI says it dedicated over 700,000 A100-equivalent GPU hours to automated red teaming for universal jailbreaks, complemented by human red-teamers. Cerebras is serving Sol at up to 750 tokens per second for select customers. Microsoft is making Sol the preferred model in Microsoft 365 Copilot as of today.

Two labs, 48 hours, one administration holding the release valve. The frontier is now something governments schedule.

Sources

• https://openai.com/index/previewing-gpt-5-6-sol/
• https://techcrunch.com/2026/07/09/openai-launches-its-new-family-of-models-with-gpt-5-6/
• https://www.cnbc.com/2026/07/08/openai-expanding-gpt-5point6-ai-model-release-ending-government-limits.html
• https://www.axios.com/2026/07/08/spacexai-grok-new-model
• https://x.ai/news/grok-4-5