GPT-5.6 Sol Pro

GPT-5.6 Sol Pro is positioned as the highest-capability option in the GPT-5.6 family, using the same underlying model as GPT-5.6 Sol with a higher reasoning setting for more demanding work. It is intended for complex coding, knowledge work, research, cybersecurity, science, computer use, and design, making it a practical choice for workflows that benefit from sustained analysis rather than a lighter, faster response mode.

Independent model reporting highlights particularly strong abstract reasoning and long-horizon agentic coding: it records 85.4% on ARC-AGI-2 at high effort, with higher scores reported at greater effort settings, alongside 88.8% on Terminal-Bench 2.1 and 64.6% on SWE-bench Pro single-agent. Together, these results support a focus on difficult programming tasks, command-line work, and research-oriented problem solving, while the source does not establish a separate training architecture, parameter count, or broad state-of-the-art claim.