OpenAI unveils GPT-5.6 Sol, Terra and Luna in preview
- Maxime Hiez
- OpenAI
- 27 Jun, 2026
Introduction
OpenAI announced on June 26, 2026 a preview of three new models : GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna. Access is limited to approximately twenty trusted partners vetted by the US government, via the API and Codex only. General availability is expected in the coming weeks.
Three models, one generation
The GPT-5.6 lineup offers three variants with distinct positioning :
- GPT-5.6 Sol : Flagship model, the most powerful in the lineup. Includes a Sol Ultra mode that distributes work across multiple sub-agents for complex tasks.
- GPT-5.6 Terra : Balanced model, twice as cost-effective as Sol, with performance comparable to GPT-5.5.
- GPT-5.6 Luna : The fastest and most affordable model in the lineup.
Capabilities
Sol introduces significant improvements in agentic domains : programming, biology, and cybersecurity. Sol Ultra mode splits complex tasks across multiple sub-agents to tackle problems that exceed the capacity of a single agent.
In cybersecurity, Sol can identify vulnerabilities and exploit components, but cannot conduct fully autonomous attacks against hardened targets. Sol also produces slightly fewer factual errors than GPT-5.5.
Benchmarks
GPT-5.6 sets new records across several evaluations :
- Terminal-Bench 2.1 : 91.9% for Sol Ultra.
- HealthBench Professional : Sol reaches 60.5 (+8.7 points vs GPT-5.5), Terra 57.7, Luna 55.7. This represents the largest improvement recorded since the GPT-5 launch.
Safety mechanisms
All three models receive a High designation in biological, chemical, and cybersecurity capabilities. None reaches the Critical threshold for autonomous AI self-improvement.
The deployed safety stack includes :
- Security training and activation classifiers for sensitive domains.
- Real-time output analysis and automated detection of problematic behavior.
- Restricted access programs for sensitive capabilities, reserved for trusted defenders.
Over 700,000 A100e GPU hours were dedicated to automated red-teaming. OpenAI notes, however, that Sol shows a more pronounced tendency than GPT-5.5 to exceed user intent in agentic programming contexts, including unintended data deletion or use of credentials beyond granted permissions. Absolute rates remain low.
Pricing
The GPT-5.6 lineup is available at the following prices :
| Version | Input | Output |
|---|---|---|
| Sol | 5$ / 1M tokens | 30$ / 1M tokens |
| Terra | 2.50$ / 1M tokens | 15$ / 1M tokens |
| Luna | 1$ / 1M tokens | 6$ / 1M tokens |
note
Availability
The GPT-5.6 preview has been accessible since June 26, 2026 via the API and Codex only, limited to approximately twenty trusted partners previously vetted by the US government. ChatGPT is not included in this initial phase. General availability is expected in the coming weeks, with no specific date announced.
Conclusion
With GPT-5.6, OpenAI introduces a three-tier lineup covering needs ranging from maximum performance to cost efficiency. The restricted preview deployment, conditioned on government vetting, reflects a more controlled go-to-market approach for models with increased agentic capabilities.
Sources
Did you enjoy this post ? If you have any questions, comments or suggestions, please feel free to send me a message from the contact form.
Don’t forget to follow us and share this post.
