OpenAI has released the GPT-5.6 family of models for general availability, introducing three new variants: Sol as the flagship, Terra for balanced workloads, and Luna for cost efficiency.
- GPT-5.6 Sol achieves state-of-the-art results on Agents’ Last Exam (53.6) and Artificial Analysis Coding Agent Index (80), outperforming Claude Fable 5 with fewer tokens and lower estimated costs.
- The family introduces "ultra" mode, which coordinates four agents in parallel to accelerate complex tasks, and Programmatic Tool Calling for efficient tool coordination.
- GPT-5.6 sets new benchmarks on BrowseComp (92.2%) and OSWorld 2.0 (62.6%), while significantly improving cybersecurity performance on ExploitBench1 (73.5%) and ExploitGym.
- Enhanced design judgment allows the models to create polished, interactive interfaces and follow complex reference formats for documents and spreadsheets more faithfully.
The release aims to provide stronger intelligence and efficiency per dollar across coding, knowledge work, and cybersecurity tasks while maintaining robust safety safeguards.