On Friday, during Day 12 of the “12 Days of OpenAI,” OpenAI CEO Sam Altman announced its newest AI “reasoning models,” o3 and o3-mini, which build on the o1 models launched earlier this year . The company is not releasing them yet, but will make these models available today for public safety testing and research access.
The models use what OpenAI calls the “private thought chain,” where the model pauses to examine its internal dialogue and plan ahead before responding, what you might call “simulated reasoning” (SR) – a form of AI that goes beyond the basic language of common language use. models (LLMs).
The company named the model family “o3” instead of “o2” to avoid potential trademark conflicts with British telecom provider O2, according to The Information. During Friday's livestream, Altman acknowledged his company's naming shortcomings, saying, “In the great tradition of OpenAI being really, really bad at names, it will be called o3.”
According to OpenAI, the o3 model achieved a record score on the ARC-AGI benchmark, a visual reasoning benchmark that has remained undefeated since its inception in 2019. In low-computation scenarios, o3 scored 75.7 percent, while in high-computation scenarios in testing it reached 87.5 percent – comparable to human performance at an 85 percent threshold.
OpenAI also reported that o3 scored 96.7 percent on the 2024 American Invitational Mathematics Exam, missing only one question. The model also achieved 87.7 percent on GPQA Diamond, which includes graduate-level biology, physics and chemistry questions. On EpochAI's Frontier Math benchmark, o3 solved 25.2 percent of problems, while no other model exceeded 2 percent.
