Sam Altman stated that an internal OpenAI model developed after Astra can resolve mathematical problems that are impossible for the world's best human mathematicians. In a conversation with Salesforce CEO Mark Benioff, Altman described a performance trajectory where GPT 5.5 mirrored an average math professor and GPT 5.6 a top 2% professor. This internal model represents a capability increase over Astra, which previously set the company's internal benchmark. The system recently solved the Navier-Stokes equations in 88 hours using 10,000 coordinating AI agents.

Recursive self-improvement is currently the top priority for OpenAI research to enable models to assist in developing their successors. Noam Brown, an OpenAI researcher, noted that while AI can now generate math results rapidly, human verification remains a bottleneck. Altman warned that open-source models are nearing a threshold where they could cause serious damage through new cyber threats, though he maintained that the development of open source should not be stopped.