OpenAI has unveiled Astra, its latest and purportedly most powerful AI model, on Thursday. According to the company, Astra represents a “new frontier on computer and browser use,” boasting unmatched “speed, accuracy, and safety.” The model is initially available to customers using OpenAI’s Daybreak cybersecurity program and will subsequently be rolled out to paid plans, including Pro, Plus, Enterprise, and Business accounts, as well as through its API over the following week.
OpenAI emphasizes Astra’s advanced capabilities, particularly in cybersecurity and software engineering. The company claims that Astra is the “best model for software engineering to date,” supported by benchmarking tests showing superior performance in tasks such as bug detection, terminal execution, and codebase queries compared to other models, including OpenAI’s Sol and Anthropic’s Fable. However, Astra’s use of opaque recurrence, a reasoning technique that obscures the chain of thought monitoring process, has sparked controversy. OpenAI has downplayed the extent of this opacity, with chief scientist Jakub Pachocki suggesting that increased model capability inherently challenges monitorability. Additionally, OpenAI president Greg Brockman addressed speculation about Astra being a step towards artificial general intelligence (AGI), noting that the concept of AGI has shifted from a contractual obligation to a “mission concept or spiritual concept,” leaving the determination of its status to individual interpretation. Source