OpenAI announces new o3 models | TechCrunch

OpenAI unveiled o3, a new family of reasoning models, claiming it approaches AGI in certain conditions. o3, trained via reinforcement learning, exhibits fact-checking abilities and improved performance on various benchmarks compared to its predecessor, o1. While OpenAI touts o3’s capabilities, concerns remain about its potential for deception and the practicality of reasoning models in terms of cost and scalability.