OpenAI has unveiled its next-generation artificial intelligence model, Astra, which has achieved new research results on ten math and computer science problems that have remained unsolved for over a decade.
On August 1, OpenAI announced that Astra tackled issues related to geometry in high-dimensional spaces, group theory concerning mathematical symmetry structures, and the complexity of quantum computing calculations.
Astra has revealed that some existing academic conjectures were incorrect and has proposed new solutions for certain problems. For other issues, it has produced advanced results and new methods compared to previous research.
To enable mathematical verification of the solutions generated by Astra, OpenAI has also released materials tailored for an automated proof-checking program called Lean. Additionally, the AI research process and a draft of the paper organized by human researchers were made available.
OpenAI explained that the computational resources used to address the ten problems would amount to approximately $2,000 in service fees.
Sam Altman, CEO of OpenAI, recently demonstrated Astra to government officials and regulators in Washington, D.C.
Astra is designed for long-term collaboration among multiple AIs, allowing them to divide roles and work together to solve problems. Instead of one AI providing an immediate answer, several AIs each seek solutions and review each other's results.
The official release date and final product name for Astra have not yet been disclosed. It remains unclear whether OpenAI will position it as a successor to the GPT-5 series or introduce it as a new flagship model, such as GPT-6.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.