OpenAI has officially released GPT-6 Astra, its next-generation frontier model, ending weeks of speculation about the launch. The company describes Astra as the most intelligent and aligned model in the world, claiming new state-of-the-art results in computer use, browsing, software engineering, cybersecurity, science and professional work. The release follows days after OpenAI warned that Astra crosses its Critical threshold for advanced cyber capabilities, part of the reason parts of the model were held back while safety controls were hardened.
Early benchmarks back up the claims. OpenAI says GPT-6 Astra sets new highs on FrontierMath Tier 4, ARC-AGI 3 and TerminalBench-4.0, the toughest public evaluations for research-level math, abstract reasoning and terminal-based agent work. Third-party testers have already started probing the model, and early results point to strong gains on agentic coding and computer-use tasks, the areas where OpenAI says Astra does the most heavy lifting.
GPT-6 Astra is rolling out now to a limited set of organizations, with ChatGPT Plus, Pro, Business and Enterprise users, plus the OpenAI API and AWS, gaining access over the coming days. For founders, the release is another signal that software is becoming the interface for AI agents: Astra is built to complete work inside real applications rather than just chat. Expect a wave of agent-ready tools and AI-first workflows as developers rebuild for a model that can operate a computer end to end.