TheModelDesk

OpenAI has released the GPT-5.6 model family, consisting of three tiers: the flagship Sol, the balanced Terra, and the cost-effective Luna. Sol establishes a new performance benchmark in cybersecurity, science, and coding, offering higher intelligence density by achieving better results with fewer tokens and lower costs compared to previous frontier models.

Technical evaluations highlight significant gains in efficiency. On the Agents’ Last Exam, Sol scored 53.6, surpassing Claude Fable 5 by 13.1 points. Additionally, Sol's max reasoning mode achieves nearly the same intelligence index as Fable 5 but completes tasks in 61% less time at approximately half the cost. In coding, Sol reached a score of 80 on the Artificial Analysis Coding Agent Index, outperforming Fable 5 while reducing output tokens and time by over half.

The update introduces a "ultra" setting for coordinating multiple agents across parallel streams and a "max" reasoning mode for complex problem-solving. A key technical addition is Programmatic Tool Calling in the Responses API, which allows models to execute lightweight programs to manage tools and filter intermediate data, reducing the need for constant developer guidance.

To ensure security, the models underwent a rigorous evaluation period involving human red teaming and automated testing. The resulting framework integrates trained model protections with real-time monitoring and risk-calibrated access controls to prevent misuse without hindering legitimate professional workflows.