GPT-6 Astra Launched: OpenAI Claims It’s the World’s Smartest, Potentially Kicking Off the AGI Era

Flash News from Insight Beating AI: OpenAI has launched GPT-6 Astra, its new flagship model focused on computer operations and long sequential tasks. The model can independently operate web pages and software, completing tasks including information retrieval, form filling, spreadsheet creation, coding, and test execution in sequence. OpenAI claims this is currently the "world's smartest and most aligned" model. In official evaluations, Astra scored 97.6% on FrontierMath Tier 4. On the ARC-AGI-3 benchmark, it achieved 99.9% when paired with OpenAI's context management mechanism, and 62.7% under a unified standard framework. It earned 57.9% on Terminal-Bench 4.0, slightly outperforming Claude Fable 5.1's 55.8%. For computer operations, Astra's score on OSWorld 2.0 rose to 72.6% from GPT-5.6 Sol's 65.7%, while simulated task time dropped from around 75 minutes to 40 minutes. When paired with the new Codex, Mind2Web task speed reached 1.9 times the previous level. OpenAI President Greg Brockman stated that looking back in a few years, people may regard Astra or this time point as the starting point of AGI. However, the "world's smartest" label remains OpenAI's own positioning. In independent testing by Artificial Analysis, Astra max has an Intelligence Index score of 61, ranking 8th currently; Claude Fable 5.1 max scores 66, still leading. Astra is also OpenAI's first model to meet the Critical cybersecurity capability threshold. During testing, it discovered and exploited two previously unknown zero-day vulnerabilities. The initial release is limited to a small number of organizations. Access will be gradually extended to Plus, Pro, Business, Enterprise, API, and AWS users in the coming days.