OpenAI, led by CEO Sam Altman, has unveiled a new series of advanced reasoning models under its o1 series. These models, according to the company, are designed to handle more complex tasks and solve difficult problems in fields like science, coding, and math. The models, named OpenAI o1 and OpenAI o1-mini, have been developed to mimic human-like problem-solving, spending more time thinking through challenges before responding.
Performance Comparable to PhD Students in Science and Math
OpenAI claims that the new o1 series models demonstrate reasoning abilities comparable to PhD students, particularly in challenging subjects like physics, chemistry, and biology. The company also noted the model’s strong performance in math and coding.
“In a qualifying exam for the International Mathematics Olympiad (IMO), GPT-4 was able to solve only 13% of the problems correctly, whereas the reasoning model from the o1 series achieved an impressive 83%,” OpenAI stated.
The o1 model is said to have broad applications across various fields. Healthcare researchers can use it to annotate cell sequencing data, while physicists may rely on it to generate complex mathematical formulas for quantum optics. Developers across different industries can leverage the model to build and execute multi-step workflows efficiently.
OpenAI has also introduced a smaller version of the model, called OpenAI o1-mini, specifically designed for coding tasks. According to the company, this version is 80% cheaper than the full o1-preview model, making it an affordable option for applications that require reasoning without the need for extensive world knowledge.
Availability of the OpenAI o1 Series
The OpenAI o1 series is being rolled out for ChatGPT Plus and Team users, with both o1-preview and o1-mini available through manual selection in the model picker. The company has set weekly limits of 30 messages for o1-preview and 50 for o1-mini but plans to increase these rates in the future. Additionally, OpenAI is working to enable ChatGPT to automatically select the appropriate model based on the prompt.
With inputs from agencies
Image Source: Multiple agencies
© Copyright 2024. All Rights Reserved Powered by Vygr Media.
