OpenAI o3
The strongest inference model of the new generation
0
- Top level mathematical reasoning ability: O3 performs excellently on complex mathematical problems, achieving an accuracy rate of 96.7% in the AIME Mathematics Competition in the United States.
- Excellent programming performance: Achieved an ELO score of 2727 on the CodeForces programming competition platform, surpassing the level of top programmers and supporting code generation and execution for complex tasks.
- Scientific problem-solving ability: In the GPQA scientific benchmark test, O3 achieved an accuracy rate of 87.7%, significantly exceeding the average level of human experts.
- Transparent reasoning path: provides a clear reasoning process that can demonstrate the logical thinking and intermediate conclusions of each step.
- Efficient multitasking: Supports long context input and can handle complex multi-step instructions.
- Lightweight version of o3Mini: Provides low-cost and efficient computing power, suitable for budget limited application scenarios.
- Powerful multimodal support: capable of handling mixed input of text and images, providing powerful support for multimodal inference scenarios.
Product Details
The OpenAI o3 model is a new generation of inference model after o1, including two versions: o3 and o3-mini. Under certain conditions, O3 approaches General Artificial Intelligence (AGI) and scores as high as 87.5% in the ARC-AGI benchmark test, far exceeding the human average. It performs well in math and programming tasks, scoring 96.7% in the 2024 American Mathematical Invitational (AIME) and 2727 in the Codeforces rating. O3 can self verify facts, reason through a "private thought chain", and improve the accuracy of answers. O3 is the first model trained using the "deliberation alignment" technique to comply with safety principles. At present, the O3 model is not widely available, but security researchers can register to preview the O3 mini model. The O3 mini version will be released at the end of January, followed shortly by the full O3 version.
