On June 26, 2026, OpenAI began a limited preview of the GPT-5.6 line — three tiers, Sol, Terra and Luna — with a new security-oversight mechanism approved by the U.S. government.
On June 26, 2026, OpenAI officially opened a limited preview of its new GPT-5.6 model line, split into three tiers for different use cases: Sol, Terra and Luna.
Three models, three purposes
- Sol: OpenAI's most powerful model to date, with clear improvements in agentic capabilities for coding, biology and cybersecurity.
- Terra: performance competitive with GPT-5.5 but about half the cost — suited to everyday work.
- Luna: optimized for high-volume calls at the lowest cost in the lineup.
On pricing (per million tokens): Sol is about $5 input / $30 output; Terra $2.5 / $15; Luna $1 / $6. All three also offer a "max" reasoning level and an "ultra" mode that coordinates many sub-agents for complex tasks.
Why can only about 20 companies use it?
The most notable point isn't the model's capability, but how it's deployed: GPT-5.6 Sol is currently available only to about 20 partners approved by the U.S. government, paired with OpenAI's strongest "safety stack" yet, with tighter protections for high-risk activity and sensitive cybersecurity-related requests.
This is a direct result of a U.S. executive order signed in early June 2026 requiring leading AI firms to voluntarily give the government early access to new models before a wide release. OpenAI said it spent weeks testing and hardening the system against real-world attacks before opening the preview.
When will it be widely available?
OpenAI said it will release all three models — Sol, Terra and Luna — widely "in the coming weeks," with no specific date. It's a clear signal that AI governance in the U.S. is shifting: from nearly no oversight to a mechanism where firms voluntarily share information early with security agencies before bringing products to market.
Comments (0)
No comments yet. Be the first to share your thoughts!