OpenAI announces its "next major model" Astra by dropping ten previously unsolved math solutions
Quick Answer
OpenAI's new model Astra has solved ten long-standing math problems, marking a significant advancement in AI-assisted research.
Quick Take
The model, which costs about $2,000 for generating solutions, formalized proofs in Lean and is expected to enhance long-term problem-solving capabilities in AI. Astra's development aligns with OpenAI's goal of creating autonomous AI researchers by 2028.
Key Points
- Astra solved ten previously unsolved math problems across various fields.
- The model's solutions cost approximately $2,000 at API rates.
- Astra formalized proofs in Lean, creating machine-checkable certificates.
- OpenAI aims for Astra to contribute to autonomous AI research by 2028.
- The model is currently undergoing testing under new US regulatory frameworks.
DeepSignal Analysis
What happened
OpenAI's Astra model has reportedly solved ten previously unsolved math problems, a significant achievement in AI-assisted research. These problems span various fields, including group theory and quantum complexity. The model's solutions were formalized in Lean, and the company claims responsibility for their accuracy.
Key evidence
- OpenAI's Astra model solved ten long-standing math problems, which had seen no progress for at least a decade.
- The solutions cover diverse areas such as high-dimensional geometry, coding theory, and group theory, with one proof resolving a major question in group theory.
- The cost to generate all ten solutions using Astra is approximately $2,000 at API rates.
Why it matters
The development of Astra aligns with OpenAI's long-term goal of creating autonomous AI researchers by 2028. This model's ability to tackle complex problems over extended periods could significantly enhance the efficiency of mathematical research. However, the implications of AI's role in research and the potential for replacing human mathematicians remain contentious topics.
What to watch
Source Excerpt
OpenAI is building a new model family called "Astra" that would let multiple agents tackle complex problems together for hours or even days. CEO Sam Altman has already demoed Astra to policymakers in Washington. OpenAI hasn't decided whether to release it as GPT-6 or a new GPT-5 variant.
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from The Decoder
See more →
An AI model programmed nonstop for 19 days on a single MirrorCode task that cost $2,600 to run
Epoch AI's MirrorCode benchmark reveals Claude Opus 4.7 as the leader with a 56% solve rate, reconstructing a 16,000-line toolkit in 14 hours. Despite this, all models tested struggle with the most complex tasks, highlighting limitations in current AI capabilities. The single task consumed $2,600 over 19 days, raising questions about cost-effectiveness in AI development.
