OpenAI Releases Astra, Its Most Capable Model Yet, Amid Debate Over AGI and Reasoning Transparency
aiinsider ·
OpenAI has released a new AI model called Astra, describing it as its most powerful and capable system yet, especially for computer and browser use. Availability began immediately for customers of OpenAI's Daybreak cybersecurity program, with broader access for Pro, Plus, Enterprise, and Business subscribers and API users rolling out over the following week. OpenAI President Greg Brockman called Astra the company's most intelligent and best-aligned model, and the firm touted benchmarks showing it outperforming OpenAI's Sol and Anthropic's Fable in bug detection and codebase analysis. The model's use of a technique called opaque recurrence, which can obscure chain-of-thought reasoning used to audit AI decision-making, drew scrutiny during the announcement. Chief Scientist Jakub Pachocki acknowledged that more capable models make monitorability harder because they can complete complex tasks with fewer or no language tokens. When asked if Astra represented the arrival of AGI, Brockman said the concept is no longer tied to a contractual definition and that he personally believes the threshold has been reached, leaving interpretation open to others.
aiinsider