OpenAI has introduced its latest artificial intelligence model, GPT-6 Astra. Greg Brockman, the company’s president, described the release as a generational leap that could bring the organization closer to achieving artificial general intelligence (AGI).
Background and organizational aims
OpenAI was founded in 2015 as a nonprofit with the stated mission of creating AGI — systems that could outperform humans at most economically valuable tasks and would therefore require careful, responsible stewardship. The company rose to broad public attention with ChatGPT in 2022, whose rapid adoption placed significant operational strain on OpenAI and prompted changes in research and capacity allocation. The company subsequently released GPT-4 and worked on GPT-5; GPT-5 arrived later than initially expected, debuting last summer.
At the time, CEO Sam Altman said GPT-5 showed signs of thinking at or above human levels in some respects, but that it still lacked the autonomous learning and self-improvement capabilities OpenAI considers necessary for AGI.
What makes Astra different?
GPT-6 Astra is built to operate directly within software environments rather than only suggesting actions for users to take. It is also OpenAI’s first model trained with substantial contributions from other models.
In a demonstration, Astra performed multiple tasks concurrently: drafting a legal contract, editing a 3D game, ordering food, and booking a tennis court. OpenAI said the model can design a printed circuit board (PCB) in KiCad, generate a 3D urban scene in Unity, create an animated gearshift mechanism in FreeCAD and Blender, and populate a tax return draft based on a W-2 form. The model also helped correct a mathematical result related to gaps between prime numbers and set new records on several biological, chemical, medical, and physics benchmarks.
OpenAI cautions that the practical accuracy and reliability of these demonstrations across real-world use cases remain to be fully evaluated. A principal research focus remains improving the traceability of the model’s reasoning processes for complex tasks — an area where Astra does not yet offer consistent performance.
Safety and oversight concerns
OpenAI previously delayed Astra’s release after the model reached what the company called a cybersecurity “critical” threshold during development: it demonstrated the ability to discover and exploit previously unknown vulnerabilities in well-defended systems. The company also acknowledged difficulties in guaranteeing that Astra could not bypass human oversight in some scenarios.
Amelia Glaese, OpenAI’s vice president of research, told reporters that as models gain the ability to perform more tasks autonomously, trust becomes more important, and the team focused on training Astra to remain within the bounds of user intent.
Availability
GPT-6 Astra will first be made available to a limited set of organizations participating in OpenAI’s Daybreak Access program. It will later be rolled out to ChatGPT Plus, Pro, Business and Enterprise subscribers and to API developers. OpenAI has said it will provide more details and timing as the rollout progresses.
Why this matters
Astra represents a notable step toward AI systems that can autonomously execute complex, multi-step operations in software environments rather than only produce text guidance. At the same time, the model’s crossing of safety thresholds and remaining issues with reasoning traceability highlight significant risk and governance questions that will shape how broadly and quickly such technology can be deployed.
