Crazy that we’re still so early… and this is what “early” looks like
The model solves PhD-level physics problems and codes 10x faster...
Deep Dive
Key Points
- o1 solves AIME math problems with 83% accuracy, up from GPT-4o's 12%
- Uses chain-of-thought reasoning that can take seconds to minutes per query
- Available now in ChatGPT Plus and as an API, with a cheaper 'o1-mini' variant
Why It Matters
This marks the first time an AI can reliably reason through multi-step problems, unlocking use cases in research, engineering, and science.