OpenAI is reportedly building Astra, a model family designed to work on problems for hours or days
Quick Answer
OpenAI is developing a new model family called 'Astra' aimed at enhancing long-duration task performance, potentially alongside existing models like Sol and Terra.
Quick Take
CEO Sam Altman showcased Astra's capabilities to coordinate multiple agents for complex problem-solving, with a long-term vision of autonomous AI research by 2028.
Key Points
- Astra aims to outperform existing models in long-running tasks like advanced math.
- The model is currently in testing and may be the first under new US AI regulations.
- OpenAI plans to showcase its current models solving ten unsolved math problems soon.
- Astra could significantly enhance human researchers' productivity by 2028.
- The development requires substantial computational resources, raising funding questions.
DeepSignal Analysis
What happened
OpenAI is developing a new model family called Astra, which aims to enhance performance on long-duration tasks. CEO Sam Altman demonstrated Astra's capabilities to coordinate multiple agents for complex problem-solving. The models are currently in testing and will be the first to undergo a new U.S. regulatory framework.
Key evidence
- Astra is designed to be more capable at long-running tasks than previous models from OpenAI, according to CEO Sam Altman.
- The models are expected to be the first tested under a new U.S. regulatory framework that requires AI models to be submitted to the federal government before public release.
- OpenAI's long-term goal is to develop a fully autonomous AI researcher by March 2028, which would rely on long-running AI processes.
Why it matters
The development of Astra reflects OpenAI's ambition to create AI systems that can handle complex, long-term tasks, which current models struggle with. This could lead to significant advancements in AI capabilities, particularly in research and problem-solving. However, the success of Astra will depend on overcoming existing limitations in multi-agent coordination and error correction.
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.
