OpenAI Launches GPT-6 Astra, Its Most Capable and Aligned Model Yet

OpenAI launched GPT-6 Astra, describing it as the world’s most intelligent and aligned model yet, combining advances in pre-training, reinforcement learning, and alignment research.

Details

• Computer use: scores 72.6% on OSWorld 2.0 versus 65.7% for GPT-5.6 Sol, completing tasks 47% faster and running 1.9x faster on Mind2Web, covering form-filling, research, code review, software installation, and troubleshooting
• Professional and academic work: 41.4% on AutomationBench (vs. 18.1%), 95.9% on BenchCAD (vs. 83.3%), 97.6% on FrontierMath Tier 4 (98% saturation), and 99.9% on ARC-AGI-3; OpenAI says it contributed to solving long-standing problems about prime number gaps
• Coding and cybersecurity: 74.1% on DeepSWE v1.1, 57.9% on Terminal-Bench 4.0 (vs. 37.3%), a perfect 100% on ExploitBench, and a new context-preservation feature for long coding sessions; during evaluation it discovered two previously unknown zero-day vulnerabilities, which were disclosed to maintainers
• Alignment gains: a 0% circumvention rate on impossible tasks, down from 48% for GPT-5.6 Sol, and three times less likely to misrepresent its own capabilities — framed by OpenAI as its most aligned model to date
• Rollout: limited organizational access starts immediately, with full availability for ChatGPT Plus, Pro, Business, and Enterprise users within days, plus access via the OpenAI API, Microsoft Azure, and AWS Bedrock; enterprise administrators must explicitly enable it, as it’s off by default at launch
• API pricing: $10 per million input tokens and $50 per million output tokens in standard mode, with a Fast Mode available at 2x speed for 2x the price, plus separate cache read/write rates; usage is included in existing subscription allowances

What happened next

GPT-6 Astra arrives as the production release of the model OpenAI previewed on September 1 in its “Path to Astra” post, which disclosed the model’s Critical-level cybersecurity capability designation and its safeguards ahead of general release. OpenAI positions the launch as outperforming Claude Fable, Claude Opus, and Gemini across most benchmarks, particularly in computer use and cybersecurity — setting up Astra as its new flagship across both consumer ChatGPT and enterprise/developer channels.