OpenAI Launches GPT-6 Astra, Its Most Capable Model Yet
New: OpenAI begins Astra's rollout today, days after Sam Altman said it crossed a 'critical' cybersecurity capability threshold.
OpenAI began rolling out GPT-6 Astra on Wednesday, calling it the most intelligent and aligned model it has built, with new state-of-the-art performance on computer use, web browsing, software engineering, cybersecurity, science and professional work tasks. It is rolling out first to a limited set of organizations before reaching ChatGPT Plus, Pro, Business and Enterprise users, plus OpenAI's API and Amazon Web Services, over the coming days.
OpenAI says the model topped FrontierMath Tier 4, ARC-AGI 3 and TerminalBench 4.0, along with Terminal-Bench Science and HealthBench Pro, and hit state-of-the-art scores on Agents' Last Exam, AutomationBench and ScreenSpot Pro - benchmarks that measure how well an AI can operate a computer to complete real workflows across professions. The company is also billing Astra as its most aligned model yet, with substantially better understanding of user intent.
The launch follows reports this week that Astra had reached 'human parity' on computer-use tasks and crossed a threshold OpenAI classifies as 'critical' for cybersecurity risk, prompting additional safety mitigations ahead of the rollout.