OpenAI is upgrading its smartest AI model with improved reasoning skills

OpenAI: today announced an improved version of its most powerful version artificial intelligence model to date, which requires more time to discuss issues, after just one day Google: announced its first model of its kind.

OpenAI’s new model, called o3, replaces o1, which the company presented in September. Like o1, the new model spends time thinking about the problem to better answer questions that require step-by-step reasoning.

The O3 model scores much higher on several metrics than its predecessor, OpenAI says, including those that measure complex coding skills and advanced math and science abilities o1 by answering the questions posed ARC-AGIa benchmark designed to test the ability of AI models to reason about problems they encounter for the first time.

Google is pursuing a similar line of research, Google researcher Noam Shazir said yesterday revealed in X’s post that the company has developed its own reasoning model called Gemini 2.0 Flash Thinking, which Google CEO Sundar Pichai called “our most thoughtful model yet.” own post.

The two dueling models show that the competition between OpenAI and Google is fiercer than ever. It is crucial for OpenAI to show that it can continue to make progress as it seeks to attract more investment and building a profitable business.Google, meanwhile, is desperate to demonstrate that it remains at the forefront of AI research.

The new models also show how AI companies are increasingly looking beyond simply scaling AI models to extracting greater intelligence from them.

Large language models can answer many questions very well, but they often falter when asked to solve puzzles that require basic math or logic. OpenAI’s o1 includes step-by-step problem-solving training, which is weak enables the AI ​​model to better solve these types of problems.

Models that reason about problems will also be important as companies look to deploy so-called AI agents that can reliably figure out how to solve complex problems on behalf of users. In Bench, a test that measures the agent abilities of models.

While a true breakthrough moment eluded the tech giants at the end of the year, the pace of AI announcements has been dizzying of late.

Earlier this month Google announced a new version of its flagship model, called Gemini 2.0, and showed it as a web browsing assistant and as an assistant that sees the world through a smartphone or smart glasses.

OpenAI has made a number of announcements ahead of Christmas, including a new version of its video generation model, a free version of its ChatGPT-powered search engine, and a way to call ChatGPT over the phone. 1-800-ChatGPT.