Astra benchmark roundup: 98.6% on ARC-AGI-3, 72.6% on OSWorld
A roundup of GPT-6 Astra's benchmark scores, with the takeaway that the focus is shifting from answering questions to completing tasks as AI agents become more real.
GPT-6 Astra just dropped. And this feels less like a better chatbot and more like a serious step toward AI that can actually do the work. Some numbers that stood out: → 98.6% on ARC-AGI-3 → 97.6% on FrontierMath Tier 4 → 95.9% on BenchCAD → 64.6% on Terminal-Bench Science → 72.6% on OSWorld V2-Offline → 74.1% on DeepSWE v1.1 for agentic coding It can also operate software, browse the web, write code, work with spreadsheets, perform research and handle complex multi-step workflows. The interesting part isn't just the benchmark scores. It's the shift from: Answer my question → Complete the task. AI agents are getting VERY real. 👀
This post quotes another post: “GPT-6 Astra announced with a claim to do anything on a computer”