OpenAI CEO Sam Altman announced the launch of two new AI models, O3 and O3-Mini, during the final day of the "OpenAI 12-day Event." These models build on the earlier O1 model and introduce a concept called "Simulated Reasoning" (SR), which enhances AI capabilities beyond basic large language models (LLMs). The models will be available for public safety testing and research.
To avoid trademark conflicts with British telecom operator O2, the models are named O3. The O3 model achieved record-breaking scores on the ARC-AGI benchmark, surpassing human-level performance in high-computation scenarios. It also excelled in the 2024 American Mathematics Competition and other advanced benchmarks.
The O3-Mini version features adaptive thinking time for varying processing speeds. It outperformed its predecessor O1 in Codeforces benchmark tests. OpenAI plans to release these models to safety researchers first, with a public launch expected soon.
