GPT-6 Astra is positioned as OpenAI's flagship model for delegation rather than chat, built around computer use and long-running professional work. It originated as OpenAI's largest training run to date, leveraging over 100,000 GPUs at the Stargate site in Texas, and notably became the first OpenAI release where earlier OpenAI models supervised the training of the new one. The model is designed to drive desktops, fill forms, build slide decks and spreadsheets, and carry out multi-step workflows, with a reported 1M-token context window that retains searchable notes across context-window rollovers in Codex.

On OpenAI's published benchmarks, Astra sets the highest scores the company has reported on abstract reasoning, math, and cybersecurity, and it is the first model to hit the Critical cybersecurity threshold under OpenAI's Preparedness Framework, prompting a phased rollout with advanced cyber capabilities gated behind the Daybreak program. Independent results temper the picture: Meta's Muse Spark 1.3 edges Astra on agentic coding, and the Artificial Analysis index has Claude ahead, reflecting that "most intelligent" is a claim measured against OpenAI's own tables rather than a consensus ranking. The practical fit is for teams delegating real agentic work who need reliable task completion, professional-grade reasoning, and strong long-context behavior.