OpenAI has introduced a new series of AI models, called OpenAI o1, specifically designed to tackle advanced reasoning tasks. These models can solve more complex problems than their predecessors, such as GPT-4, by spending more time thinking before generating responses. With the aim of improving performance in areas such as science, mathematics, and programming, OpenAI o1 promises a significant breakthrough in AI's ability to address difficult challenges.

Among the innovations in this series is the o1-mini version, a smaller and more affordable option. This model, 80% cheaper than o1-preview, is ideal for developers looking for an effective solution for tasks requiring reasoning, without needing extensive general knowledge. Both models are available to ChatGPT Plus and Teams users, with initial limits on the number of weekly messages, although these restrictions are expected to increase in the future.

Reasoning Capabilities

What sets OpenAI o1 apart is its focus on reasoning tasks. Through specialized training, the models in this series think more strategically before responding, allowing them to avoid common mistakes. In a practical example, a comparative video showed how GPT-4 and o1-preview tackled the task of counting how many times the letter 'R' appears in the word 'strawberry.' While GPT-4 responded incorrectly (two 'R's), o1-preview took a deeper approach, reviewed its response, and ultimately provided the correct answer of three 'R's. This more analytical approach avoids mistakes even in seemingly simple problems, showcasing the model's reasoning potential.

Safety and Alignment

OpenAI has also implemented significant safety improvements for these models. Thanks to a training approach that allows models to reason about safety rules in context, OpenAI o1 has proven to be much more resistant to manipulation attempts (known as 'jailbreaking'). In rigorous testing, o1-preview scored significantly higher than its predecessors in adhering to safety guidelines, reinforcing its reliability in critical applications.

Additionally, OpenAI has formalized agreements with AI safety institutes in the U.S. and the U.K., giving them early access to research versions of the model for testing and evaluations prior to public release.

Accessibility and Future

Currently, the OpenAI o1 and o1-mini models are available to ChatGPT Plus and ChatGPT Enterprise users, with plans to make o1-mini available to all ChatGPT users in the future. It is also possible to access these models via the OpenAI API, although some advanced features such as function calls, system messages, and streaming are not available in this first version.

As OpenAI continues to update the o1 series, additional features such as web browsing, file uploading, and image uploading are expected to be integrated, further expanding its capabilities for users worldwide.

You can watch the example video [here](Read More which illustrates the differences between GPT-4 and OpenAI o1-preview in a simple reasoning exercise.