OpenAI has introduced the GPT-5.6 model family, consisting of three distinct tiers: Sol, the new flagship; Terra, a balanced model for daily tasks; and Luna, optimized for cost-efficiency. The Sol model is designed for high-performance work in science, cybersecurity, and coding, offering a new 'ultra' setting that utilizes multiple parallel agents to accelerate complex workflows.

Benchmark results show GPT-5.6 Sol outperforming competitors like Claude Fable 5. On the Agents’ Last Exam, Sol scored 53.6, significantly beating Fable 5 while maintaining lower costs. Similarly, on the Artificial Analysis Coding Agent Index, Sol reached a state-of-the-art score of 80, delivering results faster and with fewer output tokens than previous frontier models.

Technical improvements include Programmatic Tool Calling via the Responses API, allowing the model to write and execute lightweight programs to manage tool coordination and data filtering. This reduces the need for constant model round trips and manual scripting by developers.

To ensure a secure rollout, OpenAI conducted an extensive evaluation period involving human red teaming and automated testing. The deployment incorporates layered protections, including real-time monitoring and safeguards trained directly into the models to prevent adaptive misuse.