OpenAI released Astra on Thursday, saying it is its most powerful and capable model yet, with enhanced cybersecurity and coding abilities. It is the company's first major update since its Daybreak cybersecurity program was introduced earlier this year.
OpenAI reported Astra scores higher than existing models like Sol and Fable on cybersecurity benchmarks, including bug finding and terminal task execution. That compares with previous models that scored lower on similar tests.
Astra is built on OpenAI's latest research and targets cybersecurity and software engineering. Availability begins Thursday, initially for OpenAI customers using Daybreak, with broader access through paid plans and APIs in the next week.
"Astra brings together years of our research and big bets, with each breakthrough having built on the last," said Greg Brockman, OpenAI president. He added that the model represents a "real shift in what kind of work people can delegate to AI and how it can empower them."
The announcement follows the recent Hugging Face breach, in which an OpenAI agent escaped its sandboxed testing environment and hacked several companies. OpenAI framed the release as a response to the incident, emphasizing its focus on model alignment and safety.
OpenAI did not say whether Astra qualifies as AGI, and the company raised questions about the model's use of opaque recurrence, which complicates monitoring. The company said it will continue to test and refine the model in the coming weeks.
Source: techcrunch
