Christmas came a couple of days early for OpenAI after its latest model achieved a breakthrough on a major AI test.
What happened: OpenAI’s o3 model, which is expected to be released publicly next year, achieved a new high score on the ARC Challenge, a test designed to be a proxy for measuring how close an AI model is to achieving artificial general intelligence (AGI).
- The creator of the test called o3’s performance a “significant leap forward in AI's ability to adapt to novel tasks.”
Catch up: The ARC Challenge tests how well AI models can solve visual reasoning puzzles that are easy for humans but — at least until now — extremely difficult for computers.
- The average human scores 84% on the challenge and o3 scored 75% — still behind typical human performance but well above where earlier OpenAI models ranked (GPT-3, for example, scored a big fat zero).
Yes, but: The people behind the ARC test caution that o3 still failed to solve many puzzles that humans are able to do easily and its relatively strong performance does not mean that AGI has been achieved.
Elsewhere: OpenAI appears to be struggling to improve its other models — GPT-5 is months behind schedule and continues to fall short of expectations, according to the Wall Street Journal.—TS