OpenAI's GPT-6 Astra on ARC-AGI-3

OpenAI's GPT-6 Astra achieved state-of-the-art performance on the ARC-AGI-3 benchmark, scoring 62.7% on the semi-private test set while using fewer actions than 96% of tested humans, demonstrating superior action efficiency. The model excelled at converting unfamiliar environments into compact symbolic world models by representing game mechanics as logical rules and developing domain-specific shorthand for planning. ARC-AGI-3 is designed to measure agentic intelligence—including exploration, modeling, goal-setting, and planning—as part of the ongoing effort to quantify the gap between current AI systems and artificial general intelligence.