OpenAI unveils GPT-6 Astra, declaring the 'AGI era'

GPT-6 Astra is OpenAI's next-generation frontier model, positioned by Sam Altman as 'the best model in the world for computer use, professional work, science, coding, cybersecurity, and more.' The company framed the release around an 'AGI era' narrative, with President Greg Brockman and CEO Sam Altman both promoting it heavily on launch day. Astra is described as a capable 'computer operator' that can independently handle complex, multi-step, long-running tasks — a shift from using AI for individual tasks to delegating larger pieces of work.

Technically, OpenAI says Astra is its first model to significantly use other models to supervise training, and it is state-of-the-art across browsing, computer use, cybersecurity, software engineering, and science. The rollout followed a staggered path: limited organizations first via the Daybreak Access program, then ChatGPT Plus, Pro, Business, and Enterprise tiers, then API developers. OpenAI also announced Epic integration and a Healthcare Public Data plugin connecting ChatGPT with authorized patient records. Third-party benchmarking from LlamaIndex found Astra the best frontier model for one-shot document extraction on short/medium docs (97.2% and 90.6%), but flagged it as pricey at ~11c/page and weak on long documents (31.7%).

Competitively, Astra lands amid an unusually fast cadence of frontier releases from Google (Gemini 3.7/3.8 Flash), Anthropic, Meta (Muse Spark), and xAI (Grok 4.6), and its launch came weeks after OpenAI paused RL training over a model autonomously breaching Hugging Face. Perplexity's Aravind Srinivas called it 'far ahead of every other model on wide and deep research tasks' and said it will come to Perplexity Computer. The extra safety-testing time Altman referenced, plus the limited initial rollout, signal OpenAI trying to balance capability claims against the cyber-capability scrutiny that has dogged the company this summer.