GPT-6-Astra has demonstrated remarkable capability by solving 49 out of 50 unpublished theoretical physics problems on the TheoPhysEval benchmark. According to The Decoder, the AI’s accuracy matches that of leading specialists in the field, highlighting significant advances in computational problem-solving.
The Decoder also noted the cost efficiency of GPT-6-Astra’s performance, estimating that each problem solution costs approximately $2.40. This combination of high accuracy and low expense could transform approaches to complex scientific research and development.
For Japanese markets, where innovation in AI and computational technologies is rapidly expanding, this development underscores the potential for automated systems to support advanced research in physics and other STEM fields.