OpenAI’s o3 Reaches Human-Level AI on ARC-AGI Benchmark

OpenAI’s new AI model, o3, has scored an impressive 85% on the ARC-AGI benchmark, matching average human performance and far surpassing the previous AI best of 55%. Demonstrating advanced adaptability and learning from limited examples, o3 marks a significant stride toward AGI. However, with task costs soaring to thousands of dollars compared to humans’ ~$5, debates spark in the AI community. Excitement and skepticism coexist as experts contemplate o3’s potential and future implications.