OpenAI Unexpectedly Releases o1, Its First Model With ‘Reasoning’ Capabilities to Solve Difficult Problems

  • The company led by Sam Altman says the model will be able to recognize its own mistakes.

  • Additionally, it’ll reportedly take time to think through answers.

The rumors were true: OpenAI was working on an artificial intelligence model with advanced capabilities. The mystery ended on Wednesday when the company led by Sam Altman finally announced the release of o1-preview, a project known internally as “Strawberry.”

This is the first preliminary version of a series of AI models whose operation is quite different from anything the industry has produced so far. Users who interact with models such as GPT-3.5 or GPT-4 find that they try to respond as quickly as possible. Notably, o1 will take some time before providing a response.

An AI Model Designed to Solve Difficult Problems

OpenAI has chosen some interesting words to describe the capabilities of its new model. The company claims that the o1 model will “think” before it responds, “much like a person would.” OpenAI also states that it’ll improve its thinking over time and learn from its mistakes.

As rumored, this model may not be suitable for all users, but it represents a significant advancement in the company’s pursuit of human-like intelligence. With o1, OpenAI expects to outperform previous science models.

For instance, users will reportedly be able to expect better answers in programming, mathematics, physics, chemistry, and biology. However, a trade-off is that it won’t be as fast as its predecessors because it’ll require time to “think” before providing answers.

As such, this model won’t be useful for users who need fast responses, such as those needed to power customer service chatbots. However, it can make a difference in academic tasks where data accuracy is crucial, addressing one of AI’s main weaknesses.

According to the company’s website, “o1 can be used by healthcare researchers to annotate cell sequencing data, by physicists to generate complicated mathematical formulas needed for quantum optics, and by developers in all fields to build and execute multi-step workflows.”

OpenAI puts an emphasis on the model’s reasoning abilities. It states that in a qualifying exam for the International Mathematical Olympiad, four-month-old GPT-4o correctly solved 13% of the problems, while o1 solved 83% of the problems.

How Can You Access OpenAI o1?

The company’s new AI model is now available in ChatGPT. However, it’s currently in a preliminary version, meaning that it’s actively being developed. At the same time, this model will only be available as a paid option.

Currently, the new AI model is accessible to ChatGPT Plus and Team users. As such, Enterprise and Edu users of ChatGPT will have to wait until next week to use the o1 model.

Users will be able to select the model they want from the ChatGPT drop-down menu. Additionally, they’ll find OpenAI o1-mini, a faster reasoning model designed primarily for programming tasks. Notably, none of the o1 models will have the ability to browse the web or analyze images to provide answers.

Moreover, both the o1 and o1-mini models will also be accessible through OpenAI’s API, enabling developers to integrate these models into their apps. Naturally, the pricing differs for these models. The o1 model is priced at $60 for 1 million output tokens, while the o1-mini model is priced at $12 for 1 million output tokens.

Image | OpenAI

Related | We Thought ChatGPT Was Great for Programming. A New Study Finds That Half of Its Answers Are Wrong

See all comments on https://www.xatakaon.com

SEE 0 Comment

Cover of Xataka On