OpenAI Unveils o3 Models: 87.5% on AGI Benchmark
Discover OpenAI's groundbreaking o3 models: redefining AI benchmarks with 87.5% AGI accuracy, innovative reasoning, and next-gen safety alignment.
OpenAI's 12-day release event is coming to a close. As expected, the new reasoning series models—o3 and o3-mini—have taken center stage as the final announcements of the event.
o3 is the successor to the o1 series models. These models are characterized by spending more time on reasoning before answering questions, which improves the accuracy of their responses. However, OpenAI skipped the naming of o2.
According to The Information, this was done to avoid copyright issues, as there is a telecommunications company named O2 in the UK, which could cause confusion. Sam Altman confirmed this during a live broadcast this afternoon.
In fact, OpenAI started to warm up the model yesterday, and developers have already found references to o3_min_safety_test on OpenAI's website.
Some were eager to see what o3 could actually do, and whether it could compete with Google's Gemini 2.0 Flash Thinking released yesterday.
