OpenAI's o3 artificial intelligence model has achieved a significant score on the ARC Challenge, a test for AI reasoning skills, leading some enthusiasts to wonder if it has reached artificial general intelligence (AGI). However, organizers of the ARC Challenge clarify that although o3 reached a milestone, it hasn't won the competition's grand prize and has not yet achieved AGI, which implies human-like intelligence. The o3 model, a successor of large language models like ChatGPT, performed on tasks designed to test general intelligence through pattern recognition in colored grids. The ARC Challenge requires computational power restrictions to prevent solving the puzzles by brute force. OpenAI's model scored 75. 7% officially, complying with the competition's limit of expenses, but did not meet the stricter private test criteria, which determines grand prize winners. Unofficially, o3 achieved an 87. 5% score by using much more computing power, with costs going up to thousands per task—much higher than the competition allows. Despite surpassing the typical human score of 84%, AGI has not been achieved, as asserted by challenge organizers and AI experts. The model also struggled to solve over 100 tasks even with substantial computing power. AI researchers, such as François Chollet of Google, highlight that solving tasks through sheer computation undermines the intention of indicating AGI.

Chollet and other experts outline that true AGI would eliminate the challenge of creating tasks that are simple for humans yet difficult for AI. Currently, o3's achievement signifies progress but not AGI. The tech industry continues to grapple with the recent slowdown in AI advancement compared to earlier explosive developments. The possibility that AI models could soon beat competition benchmarks remains, with some already scoring above 81% in evaluations. Future steps include a second, harder set of tests anticipated for 2025. The ultimate goal is someone achieving and open-sourcing the grand prize-winning solution.