OpenAI Debuts o1, Its First AI Model Built for Complex Reasoning

OpenAI has introduced its latest AI model series, o1, designed to improve AI's ability to solve complex problems in fields like math, science, and coding. If you've been following the rumors, this is the much hyped Strawberry model.

The o1 model series, which includes the high-performance o1-preview and the cost-effective o1-mini, represents a major step toward enhancing AI's reasoning capabilities, moving beyond the pattern recognition approach of earlier models like GPT-4o.

What Sets o1 Apart

Unlike previous models, the o1 series uses reinforcement learning, a method that enables the AI to approach problems step-by-step, similar to human reasoning. This approach allows the model to refine its thinking, try different strategies, and learn from mistakes. According to OpenAI, o1 vastly outperforms GPT-4o in areas like mathematics and coding.

In a qualifying exam for the International Mathematics Olympiad, o1 correctly solved 83% of problems compared to GPT-4o's 13%. It also performed in the 89th percentile in coding competitions.

However, while o1 excels at solving complex problems, it lacks some of the broader features of GPT-4o, such as web browsing and file processing. GPT-4o remains more versatile for everyday applications, but o1 is the stronger model for tasks that require deep reasoning.

Costs and Access

The o1-preview is priced higher than GPT-4o, at $15 per million input tokens and $60 per million output tokens. To make it more accessible, OpenAI has also released o1-mini, which is 80% cheaper and optimized for coding tasks.

Both models are available to ChatGPT Plus and Team users now, with broader availability planned for Enterprise, Educational, and free-tier users.

Improved Safety and Future Plans

OpenAI has incorporated advanced safety protocols into the o1 model, improving its ability to adhere to safety rules compared to earlier models. On OpenAI's jailbreaking tests, o1-preview scored significantly higher in resisting security bypass attempts.

OpenAI plans to continue updating the o1 series, adding features like web browsing and file uploading. The company will also continue developing its GPT series, aiming to push AI closer to human-like intelligence.

Source