ChatGPT Gets New o1 Model, First to Have ‘Reasoning’ for Hard Problems [CNET]

The new model is designed to solve complex problems across a small handful of fields, but OpenAI says the model performs similarly to PhD students in those tasks.

ChatGPT has a new model named o1 that’s trained to solve harder problems, analyze its answers, try different strategies and refine its thinking, OpenAI said in a blog post on Thursday. 

The new model, currently split between o1-preview and o1-mini, ranks in the 89th percentile in Codeforces’ competitive programming contests, places among the top 500 students in the US for the Math Olympiad and “exceeds PhD-level accuracy on a benchmark of physics, biology and chemistry problems,” according to OpenAI.  

“We have noticed that this model hallucinates less,” says Jerry Tworek, OpenAI’s research lead in an interview with The Verge. It’s been trained on a new optimization algorithm with a tailor-made training dataset. While past models aimed to mimic patterns in their training data, o1 uses reinforcement learning, which teaches it through rewards and penalties. 

The thing that differentiates o1 from past models is its ability to “think,” according to a report from The Information on Tuesday. This means the model doesn’t immediately begin spitting out responses and can take from 10 to 20 seconds to put together a thoughtful answer. The o1 model, which has also been referred to as “Strawberry” by onlookers (a possible reference to the viral trend of influencers asking AIs to answer how many “Rs” are in the word “strawberry”), removes the need for “chain-of-thought prompting,” where users have to ask extra questions of an AI to see its intermediate reasoning. Instead, the model is designed to show its reasoning by default.

Because o1 is still in its preview stage, there are some major limitations. Unlike GPT-4o, o1 isn’t connected to the web, can’t be used with file uploads and has a multitude of API limitations for developers. The o1-mini model differs in that it focuses on delivering fast answers to STEM-related questions. 

Competition in the AI space continues to get more fierce as every player in the big tech space aims to out-compete one another and create “agentive” AIs that can complete tasks for you. Earlier this year at Google I/O, the search giant unveiled a more powerful version of Gemini that can more naturally converse with you, even allowing you to interrupt it mid-sentence. And at the iPhone 16 launch event earlier this week, Apple bumped up the processing power of its latest handsets to be able to handle Apple Intelligence, a suite of AI features for iPhones backed with OpenAI tech.

While AI hype has been driving tech stocks to record numbers in the last two years, it seems that investors might be growing more cautious. Nvidia, the chip maker that’s creating the brains powering many of the world’s top AI data centers, saw a 10% drop last week. The tech world broadly could be cooling on AI as it waits for more concrete results from services, although that hasn’t stopped OpenAI from reaching a staggering $150 billion valuation.

For ChatGPT Plus and Team users, the o1-preview model is rolling out now. ChatGPT Enterprise and Edu users will gain access next week. Developers can also use the API for prototyping.