OpenAI has released its latest model, GPT-6 Astra. This model provides state-of-the-art performance across various fields, including computer use, browsing, software engineering, cybersecurity, science, and professional tasks.
Model ReleasesOpenAIGPT-6 Astra
OpenAI Releases Next-Generation Model GPT-6 Astra, Demonstrating High Performance in Computer Use and Mathematics
This article is a translation. Read the Japanese original
GPT-6 Astra recorded a score of 98% on FrontierMath Tier 4, a mathematics benchmark, and achieved 99.9% on ARC-AGI-3. In the field of computer use, it demonstrated high efficiency in OSWorld 2.0 simulations, reducing task completion time by approximately 47% compared to GPT-5.6 Sol. It also possesses advanced capabilities in cybersecurity, securing a score of 100% on ExploitBench.
The model's alignment capabilities have also been enhanced, allowing it to understand user intent and respect task boundaries. According to OpenAI, if Astra encounters ambiguity in interpreting instructions, it can make more accurate decisions by asking appropriate clarifying questions.
GPT-6 Astra will be rolled out sequentially to ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API, Microsoft Azure, and AWS Bedrock.
Sources
- GPT-6 Astra, Looped Transformers, and Hidden Reasoning （Hacker News Frontpage、2026-09-09）
- OpenAI 公式ブログ