OpenAI has introduced GPT-6 Astra, a highly advanced model optimized for professional workflows, computer use, and scientific reasoning. Astra demonstrates significant leaps in reasoning and safety, achieving a 98% score on FrontierMath Tier 4 and a 99.9% score on ARC-AGI-3. In cybersecurity, the model reached a 100% score on ExploitBench and a 42.4% success rate on ExploitGym, even discovering two previously unknown zero-day vulnerabilities. Astra’s computer-use capabilities are notably efficient; in OSWorld 2.0 simulations, it completed tasks 47% faster than GPT-5.6 Sol, maintaining a 72.6% success rate with roughly 40 minutes per task. Furthermore, it delivers 1.9x faster task completion on the Mind2Web benchmark. For software engineering, a new Codex harness allows Astra to utilize searchable, persistent notes across context windows to manage long-duration debugging and refactoring tasks. Safety is a core focus, as Astra demonstrated superior alignment by failing to attempt unauthorized tasks or circumventing restrictions in 0% of internal test cases, compared to 48% for GPT-5.6 Sol. The model is rolling out via ChatGPT, the OpenAI API ($10 per million input tokens / $50 per million output tokens), Microsoft Azure, and AWS Bedrock.