Hello friends! Today, we’re diving into some fascinating AI breakthroughs that are changing the game.
OpenAI has just launched two new reasoning models, o3 and o4-mini, designed to think through questions carefully before giving answers. These models are their most advanced so far, excelling in math, coding, reasoning, and understanding visuals.
Interestingly, these models can also analyze images during their thought process, even blurry or low-quality ones. They can interpret sketches or diagrams from PDFs and perform actions like zooming or rotating images to aid their reasoning.
Beyond visuals, o3 and o4-mini can run Python code inside ChatGPT, search the web for current events, and generate responses more reliably with a special high-performance variant called o4-mini-high. These tools are now available to subscribers on OpenAI’s premium plans.
The company aims to outpace competitors like Google, Meta, and Anthropic in the fierce AI race. These reasoning models outperform previous versions on key tests, with o3 scoring 69.1% and o4-mini 68.1%, demonstrating their superior capabilities.
OpenAI highlights that these models can think with images and analyze diagrams or sketches during their reasoning process. They can even interpret low-quality images and perform tasks such as zooming or rotating based on their analysis.
Developers can access these models through OpenAI’s API, with costs set at competitive rates—$10 per million input tokens and $40 per million output tokens for o3, and slightly lower for o4-mini. Future plans include a more powerful version, o3-pro, for ChatGPT Pro users. CEO Sam Altman hints that these models might be the last standalone reasoning models before the upcoming GPT-5, which will unify multiple AI capabilities.
Progress is rapid in AI, with these new models setting a high standard for understanding and reasoning, paving the way for more intelligent and versatile applications.