OpenAI has launched its GPT-5.6 model family, featuring three distinct versions: Sol, Terra, and Luna. The flagship model, GPT-5.6 Sol, achieves state-of-the-art results across domains like coding, cybersecurity, and scientific reasoning, reportedly outperforming previous frontier models with reduced token usage and cost. This efficiency translates to better performance per dollar for users.

New capabilities include ultra, a setting that coordinates multiple agents in parallel to accelerate complex tasks. GPT-5.6 Sol demonstrates enhanced computer vision and design judgment, enabling it to inspect, refine, and deliver polished results. On the Agents' Last Exam, GPT-5.6 Sol achieved a score of 53.6, surpassing Claude Fable 5 by a significant margin. The models also introduce Programmatic Tool Calling in the Responses API, allowing lightweight programs to coordinate tools, process intermediate results, and adapt workflows, reducing the need for extensive scripting.

GPT-5.6 Sol is also OpenAI's strongest coding model to date, setting new state-of-the-art results on the Artificial Analysis Coding Agent Index. The models exhibit improved design judgment, creating tasteful and functional interfaces with high-level direction. They can infer and apply design systems from reference materials, producing more polished documents and spreadsheets. Furthermore, GPT-5.6 shows advancements in cybersecurity, achieving frontier performance on benchmarks like Exploit Bench and Exploit Gym.