OpenAI's o3 Model Scores 75.7% on ARC-AGI, Set for Early 2025 Release

Dec 21, 2024, 12:48 PM

OpenAI has unveiled its latest AI model, o3, marking a significant advancement in AI reasoning capabilities. The o3 model, along with its smaller counterpart o3-mini, is set to be released in early 2025 following safety testing and red teaming. o3 achieved a breakthrough score of 75.7% on the ARC-AGI benchmark's semi-private evaluation set, with a high-compute configuration reaching 87.5%. Despite these impressive results, experts caution that o3 does not yet constitute artificial general intelligence (AGI), as it still fails on some tasks that are straightforward for humans. OpenAI's o3 model represents a step forward in AI's ability to adapt to novel tasks, but it is not considered AGI due to its limitations in handling certain easy tasks and the high cost of operation, which can reach thousands of dollars per task.