OpenAI Unveils New Artificial Intelligence Capable of Solving Decades-Old Math Problems
San Francisco, Saturday, 1 August 2026.
OpenAI introduced its Astra model family, which uses collaborating AI agents over extended periods to solve ten decade-old mathematics problems for just $2,000, marking a major leap in scientific reasoning.
Technical Capabilities and Cost Analysis
OpenAI officially confirmed the name “Astra” for its next major model family on 2026-07-31 [1]. The system successfully solved ten open problems in mathematics and theoretical computer science that had remained unsolved for at least a decade [1]. Generating these ten solutions would cost approximately $2,000 at Sol API rates, averaging 200 per problem [2]. Noam Brown, a researcher behind the test-time reasoning technology, described the development as a major step for scientific reasoning [2].
Deployment Timeline and Strategic Shift
Prior to the official release, CEO Sam Altman demoed the system in Washington, D.C., during the week of 2026-07-27 [6]. The model is designed for long-running tasks lasting hours to days, allowing multiple agents to collaborate autonomously [6]. This shift highlights a business trend toward high-value corporate and scientific problem-solving rather than real-time chat interactions [GPT].
Distinguishing OpenAI Astra from Competitors
It is crucial to distinguish this model from Google’s Project Astra, which is a multimodal AI research prototype designed for interactive assistance [3]. Google’s prototype functions via mobile phone cameras for real-time environmental description and object identification [3]. Conversely, OpenAI’s Astra focuses on agentic workflows for complex reasoning [5].
Acquisition Context and Model Classification
Additionally, confusion may arise from OpenAI’s announcement on 2026-07-31 regarding the acquisition of Astral, a Python tooling company [4]. This transaction is subject to customary closing conditions and regulatory approval, with companies operating independently until then [4]. The Astra model family exists as a new model class alongside OpenAI’s existing Sol, Terra, and Luna families [1].
Regulatory Landscape and Future Roadmap
The Trump administration aims to finalize a new AI regulatory framework by 2026-08-02, which would require voluntary submission of models like Astra for safety reviews [1][6]. Astra is expected to be the first model tested under this framework [1]. OpenAI aims to achieve a fully autonomous AI researcher by March 2028 [1].