Hype Check: GPT-5.6 Sol Still Scores in the Low Teens on ARC-AGI Even at Max Reasoning

ARC Prize·medium signal

Against the 'new standard for intelligence' framing, ARC Prize's published GPT-5.6 results show Sol only performs at max reasoning effort and lands around 13.3% on the public and 7.8% on the semi-private ARC-AGI sets — a reminder that record scores on coding and professional-workflow benchmarks don't transfer to novel fluid-reasoning tasks where humans still hit ~100%. The builder takeaway: GPT-5.6 is a step up for agentic and coding work, not evidence that general reasoning has been solved.