The release of OpenAI’s models, 03 and 04 Mini, has excited the AI community, with experts praising 03 for its exceptional capabilities, including a Mensa IQ score of 136 and advanced problem-solving skills. While both models demonstrate significant advancements in areas like geoging and complex math problem-solving, some limitations remain, highlighting the continued value of human involvement in AI tasks.
The recent release of OpenAI’s models, 03 and 04 Mini, has generated significant excitement in the AI community. Daria Enutz, who had early access to these models, praised 03 for its exceptional capabilities, claiming it operates at or near genius level. Notably, 03 achieved a Mensa IQ score of 136, surpassing previous models like Gemini 2.5 Pro, which scored 128. Enutz highlighted 03’s ability to use tools effectively and iteratively during problem-solving, making it a remarkable advancement in AI technology.
The video also features insights from various industry experts. Amjad Msad, CEO of Replit, noted that 04 Mini can perform tool calls within its reasoning chain, enhancing its functionality. Dave Shapiro, an AI content creator, expressed that 03 represents a significant innovation in AI, comparable to the impact of ChatGPT. He emphasized that 03’s ability to tackle complex topics with precision and clarity marks a substantial improvement over earlier models.
One of the standout features of 03 is its proficiency in geoging, the ability to identify locations from random Google Street View images. An example showcased 03’s capability to pinpoint a location in Eastern Canada based on a challenging image, demonstrating its advanced reasoning skills. However, the video also cautioned that while AI may excel in tasks like geoging, human involvement remains valuable, as the enjoyment of human competition persists even in areas where AI outperforms.
The video further highlights 04 Mini’s impressive performance in solving complex math problems, outperforming human solvers in speed and accuracy. It achieved remarkable results in various coding tasks, surpassing previous models in coding intelligence. The comparison of token usage across different models indicated that 04 Mini is efficient, using fewer tokens for reasoning, which enhances its overall performance and cost-effectiveness.
Despite the advancements, the video acknowledges that not all tests yield perfect results, with some instances of failure noted. For example, 03 struggled with a visual recognition task involving color identification. Overall, the release of 03 and 04 Mini represents a significant leap in AI capabilities, with many in the industry eager to explore their potential. The video concludes by inviting viewers to share their experiences with the new models and encourages engagement with the content.