Photo by Ash Valiente on Pexels

A new model, dubbed GPT-5 Mini (High), has reportedly achieved impressive scores on the ARC-AGI benchmark, surpassing the performance of established models like Gemini 2.5 Pro and Claude Opus 4. Sources indicate that GPT-5 Mini not only achieved higher accuracy but also did so at a considerably lower cost. This development points towards significant progress in the efficient design and optimization of advanced AI models. Further details can be found in the original Reddit post: [https://old.reddit.com/r/artificial/comments/1mknlss/gpt5_mini_quietly_outperforms_gemini_25_pro/]