July 9, 2026. GPT-5.6 goes public today. OpenAI confirmed in a Tuesday evening post on X that GPT-5.6 Sol, Terra and Luna launch publicly this Thursday, adding that it is "expanding preview access globally now." The release ends a 13-day holding pattern: the models debuted June 26 to roughly 20 trusted partner organizations at the US government's request, with broad access gated behind a federal safety review, as Nextgov and Engadget report.
Key developments
- All three tiers go public July 9 across ChatGPT, Codex and the API, per OpenAI. Sol is the flagship, Terra the everyday workhorse, Luna the fast low-cost tier.
- The government gate opened after additional evaluations. Engadget reports the administration cleared the wider release after further testing and meetings under June's voluntary-review executive order. Whether any weights or parameters changed during the review is undisclosed; OpenAI did not answer Nextgov's question on it.
- The naming system is doing real work. The number marks the generation while Sol, Terra and Luna are durable capability tiers that advance on their own cadence, so expect tier-level upgrades between generations.
- June-announced API pricing stands: Sol at 5 dollars input and 30 dollars output per million tokens, Terra at 2.50 and 15, Luna at 1 and 6. OpenAI says Terra performs comparably to GPT-5.5 at half the price, and Sol, tuned for biology, chemistry and cybersecurity, ships with a stronger Sol Ultra variant.
- Geographic scope is still firming up. OpenAI's post says preview access is expanding globally, though 9to5Mac notes it was not yet confirmed whether today's public release reaches every region at once. Sam Altman said in June the company was working toward a global release.
What it means for operators
The action item is the routing math. If Terra really matches GPT-5.5 at half the price, every workload currently on GPT-5.5 deserves a re-test this week, and Luna resets the floor for high-volume extraction, classification and support tasks that were drifting toward cut-price open-weight models, the cost pressure we covered in yesterday's Chinese-models brief. Run your own evals before switching, exactly as we advised when Sonnet 5 reset the mid-market: a vendor benchmark is a hypothesis, your production traffic is the test. Budget an afternoon to replay your top three prompts and one agentic workflow against Terra and Luna, compare cost per completed task rather than cost per token, and only then move traffic.
And bank the policy lesson from the gated June launch: frontier access is now a policy variable that can close and reopen in weeks, on the government's calendar rather than the vendor's. Keep your automation stack model-agnostic behind a routing layer, and let an AI engineer wire that layer once instead of rewriting integrations every time the frontier moves.
Frequently Asked Questions
They are the three tiers of OpenAI's GPT-5.6 family, public as of July 9, 2026. Sol is the flagship for the hardest problems, Terra is the balanced everyday model that OpenAI says matches GPT-5.5 at half the price, and Luna is the fast, lowest-cost tier.
At the US government's request, OpenAI initially limited GPT-5.6 to about 20 trusted partners on June 26, 2026 while federal partners ran safety evaluations focused on the models' strong biology, chemistry and cybersecurity capabilities. The wider release was cleared 13 days later.
Per the June announcement, API pricing per million tokens is 5 dollars input and 30 dollars output for Sol, 2.50 and 15 for Terra, and 1 and 6 for Luna. ChatGPT access rolls out across paid plans starting today.
Test first. Terra's price-performance claim makes it the obvious candidate to re-benchmark against GPT-5.5 on your own workloads, and Luna is worth testing for high-volume tasks. Keep routing model-agnostic so you can adopt wins without rearchitecting.