GPT-6 Astra Release: OpenAI Claims World's Smartest, AGI Era May Begin Here

Dynamic Beating AI News Flash: OpenAI has released GPT-6 Astra, the new flagship model focusing on computer operation and long tasks. It can operate web pages and software by itself, consecutively completing tasks such as information retrieval, form filling, spreadsheet manipulation, coding, and running tests. OpenAI claims that this is currently the "smartest, most aligned" model in the world.

In official evaluations, Astra scored 97.6% in FrontierMath Tier 4. In the ARC-AGI-3 validation test, it achieved 99.9% with OpenAI's context management mechanism, and 62.7% under a unified standard framework. In Terminal-Bench 4.0, it scored 57.9%, slightly higher than Claude Fable 5.1's 55.8%. In terms of computer operation, OSWorld 2.0 improved from 65.7% in GPT-5.6 Sol to 72.6%, with the simulation task time reduced from about 75 minutes to 40 minutes. With the new version of Codex, the speed of Mind2Web tasks has increased by 1.9 times compared to before.

OpenAI CEO Greg Brockman mentioned that looking back in a few years, people might consider Astra or this time point as the starting point of AGI. However, being the "world's smartest" is still OpenAI's own positioning. In an independent test by Artificial Analysis, Astra's max Intelligence Index is 61, ranking 8th; while Claude Fable 5.1 has a max of 66, maintaining its lead.

Astra is also OpenAI's first model to reach the Critical network security capability threshold. During testing, it also discovered and exploited two previously unknown zero-day vulnerabilities. The initial rollout is limited to a few organizations. Access will gradually open up to Plus, Pro, Business, Enterprise, API, and AWS in the coming days.