Search
OpenAI hosted a technical livestream at 1 AM today, revealing its groundbreaking multimodal models, o4-mini and the robust o3. These models are capable of simultaneously processing text, images, and audio, and they possess agent capabilities that facilitate the automatic utilization of tools like web search, image generation, code parsing, and deep-thinking modes. According to test data shared by OpenAI, o4-mini attained impressive accuracy rates of 93.4% and 92.7% in the AIME benchmarks for 2024 and 2025, respectively, outperforming the full-strength o3 and solidifying its position as the model with the highest accuracy rate globally. Furthermore, in the Codeforces test, o4-mini scored 2700 points, placing it among the top 200 most skilled programmers worldwide.
-
C114 Communication Network -
Communication Home
