Introducing OpenAI's O1-Preview-2024-09-12: The Next Leap in AI Reasoning
OpenAI has unveiled a groundbreaking series of AI models, the O1, designed to enhance reasoning capabilities and tackle complex tasks more effectively. The O1-preview and O1-mini models are now available starting September 12, 2024, for ChatGPT Plus and Team subscribers, with ChatGPT Enterprise and Edu users gaining access next week. Developers with API usage tier 5 can also start prototyping with both models, subject to a rate limit of 20 RPM. Plans are underway to extend access to O1-mini for all ChatGPT Free users.
The O1 models excel in complex reasoning tasks, particularly in science, coding, and math. They have demonstrated significant performance improvements, scoring 83% in the International Mathematics Olympiad (IMO) qualifying exam, compared to GPT-4o's 13%, and ranking in the 89th percentile in coding competitions like Codeforces. On challenging benchmark tasks in physics, chemistry, and biology, they perform similarly to PhD students.
The enhanced reasoning capabilities of the O1 models stem from a new training algorithm and reinforcement learning approach, which use rewards and penalties to teach the model to think step-by-step. This method reduces the occurrence of "hallucinations" or poor outputs, ensuring more reliable responses.
While the O1 models currently lack features like web browsing and file and image uploading, these are planned for future updates. OpenAI anticipates regular updates and improvements as the models evolve.
In terms of pricing, developers can access the API at $15 per million input tokens and $60 per million output tokens for O1-preview, higher than GPT-4o. ChatGPT Plus and Team users have weekly rate limits of 30 messages for O1-preview and 50 messages for O1-mini.
OpenAI has also implemented a new safety training approach that leverages the reasoning capabilities of the O1 models to adhere to safety and alignment guidelines more effectively.
This new series of models marks a significant advancement in AI capabilities, particularly in complex reasoning tasks, representing the beginning of a new era of "reasoning" models from OpenAI.
