GPT-6 Astra pushes frontier models into delegated work
The important shift is not another chat benchmark: OpenAI says Astra can handle computer use, browsing, software engineering, cybersecurity, science, and other professional workflows with less micromanagement. It reports 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 64.6% on Terminal-Bench Science 0.1, beating GPT-5.6 Sol’s 22.4% at a lower-cost setting. The API rollout matters because these capabilities are moving directly into agents and production tools.