"OpenAI officially launches GPT-6 Astra, outperforming Claude in multiple benchmarks! OpenAI has officially unveiled its new flagship model, GPT-6 Astra, which the company describes as "the world's most intelligent and aligned model." It has set new records across various fields, including computer operations, browsing, software engineering, cybersecurity, scientific research, and professional tasks. The model will be rolled out to all ChatGPT Plus, Pro, Business, and Enterprise users in the coming days, and will also be available via the OpenAI API and AWS Bedrock. In terms of benchmarks, Astra scored 97.6% on FrontierMath Tier 4 and achieved a staggering 99.9% on ARC-AGI-3, far surpassing the single-digit scores of previous models. On the programming test Terminal-Bench 4.0, it scored 57.9%, outperforming its competitor Claude Fable 5.1 (55.8%). On the safety and alignment front, OpenAI introduced a new evaluation inspired by the Hugging Face incident to test whether the model would overstep boundaries when faced with unachievable tasks. Without production environment safeguards, Astra's overreach rate was 0%, compared to 48% for the previous generation model, GPT-5.6 Sol. Notably, Astra achieved a perfect score of 100% on the cybersecurity capability test ExploitBench. OpenAI officially recognized it as reaching the "critical" risk level under its "preparatory framework." During testing, Astra even identified two previously unknown zero-day vulnerabilities, which have since been disclosed to the relevant vendors. For safety reasons, OpenAI stated that Astra will not respond to higher-level cybersecurity requests, such as creating proof-of-concept exploits, in the API environment. #GPT6Astra #OpenAI #AI #Cybersecurity #TechNews #Innovation