Astra marks the Claude 3.7 moment for knowledge work (fundaai.substack.com)

OpenAI has unveiled Astra, a significant upgrade to its AI model, representing its first major version change since GPT-5’s release thirteen months ago. Notably, Astra is the first model trained on over 100,000 GPUs, the largest training run in OpenAI’s history, and incorporates learnings from previous models to improve its efficiency. Released on September 3, Astra is now accessible to paid ChatGPT users and through various APIs, priced at $10 per million input tokens and $50 per million output tokens. With enhanced capabilities in computer use, Astra can now navigate and operate software applications like Excel and CAD programs, significantly reducing the time needed to complete tasks compared to human workers. This breakthrough paves the way for broader AI applications beyond coding, potentially transforming knowledge work across various sectors, with an addressable market exceeding $10 trillion. By automating routine professional tasks and facilitating complex workflows, Astra could streamline operations in fields such as finance, legal, and creative industries. It also demonstrates promising advancements in mathematical and scientific research, having solved two open Erdős problems and improved existing mathematical bounds—a leap that could impact future scientific breakthroughs significantly. Astra signifies a pivotal moment in AI development, setting the stage for the near-term commercialization of AI in knowledge work, while also hinting at a future where AI could autonomously conduct scientific research.