OpenAI’s o1 Reasoning Model Officially Integrated into ChatGPT
OpenAI recently announced the integration of its advanced o1 reasoning model into the ChatGPT platform, enabling users to tackle complex problems in mathematics, science, and coding. Activated via a ‘Think’ button, the model offers two options, o1-preview and o1-mini, and is currently being rolled out to all ChatGPT paid subscribers, including Plus, Team, Enterprise, and Edu plans. Free users can also access o1-mini with limitations.
Core Capabilities of the o1 Model
The o1 model is specifically designed for complex reasoning tasks, simulating a human-like step-by-step thought process and achieving remarkable results in various international benchmarks. According to official OpenAI data, o1-preview achieved an accuracy of 74.3% in the AIME 2024 math competition, a significant leap from the previous model’s 9.3%. It also scored 78% on the GPQA science benchmark and ranked in the top 500 globally on Codeforces (Elo score of 1892). The o1-mini variant achieved 66.3% on AIME and 67.7% on GPQA, making it particularly suitable for STEM-related tasks by offering faster responses at a lower cost.
How to Use It in ChatGPT
Within ChatGPT, users can click the ‘Think’ button below a message to switch to the o1 mode. The system will display the reasoning process, including error checks and alternative approaches, enhancing transparency and accuracy. o1-preview is ideal for queries requiring deep reasoning, such as graduate-level problems, while o1-mini is optimized for speed and efficiency, perfect for programming and data analysis. OpenAI has stated that this integration requires no extra subscription and the gradual rollout will eventually cover all users.
Performance Benchmarks and Practical Applications
Cross-validation also shows that the o1 series performs exceptionally well on tests like ARC-AGI, with o1-preview reaching 85% and o1-mini 79%. Leading media outlets like TechCrunch and The Verge have confirmed that this update marks ChatGPT’s transition from simple conversation to advanced problem-solving. User feedback indicates a significant improvement in accuracy for multi-step reasoning tasks, such as solving calculus or physics simulation problems.
Gradual Rollout and Future Outlook
The rollout has begun, starting with ChatGPT Plus users and subsequently expanding to other plans. OpenAI emphasizes that the o1 model is trained using reinforcement learning to better handle uncertainty and long-chain reasoning. While the o1 API is also available to developers, the ChatGPT integration further democratizes this technology. Looking ahead, OpenAI will continue to optimize the model’s performance.
