OpenAI's o3 Achieves Human-Level on ARC-AGI Benchmark

OpenAI’s shiny new model, o3, scored an impressive 85% on the ARC-AGI benchmark as of December 2024, matching human averages and smashing the old AI record of 55%. This leap showcases o3’s knack for learning from scant examples, a big step toward AGI. However, with costs soaring from $17 to thousands per task, wallets might not cheer. François Chollet tempers the excitement, reminding us AGI remains elusive. The AI community buzzes with a mix of awe and cautious optimism.