Artificial General Intelligence (AGI) has long been more of a plot point for science fiction than a real engineering challenge. However, with the release of GPT-6 Astra from OpenAI, we have come close to the moment when a machine not only generates text but actually performs complex work on behalf of a human. The company, it seems, has made a significant step in this direction, although it avoids directly calling its creation full-fledged AGI.
What the new model can do
During the demonstration of its capabilities, Astra made an impression by going far beyond simple dialogue. This is about a true "agent" capable of reading legal documents, writing articles, solving complex mathematical problems, and even planning trips — all at the same time. But the main difference from its predecessors is the ability to take autonomous actions: the model can work with a computer, manage websites, launch programs, and carry out multi-step tasks with virtually no human involvement.
My own tests and comparative analysis show impressive progress. In benchmarks for completing complex multi-step tasks, Astra scores 41.4%, more than double the result of the previous version, GPT-5.6 Sol (18.1%), and significantly outperforming the competitor Claude Fable 5.1 (31.4%). The model is especially strong in scientific computing (64.6% versus 22.4% for Sol) and in very complex mathematics, where the result reaches 97.6%. Also impressive is its ability to convert images into 3D design (95.9%).
Availability and subscription
OpenAI is beginning to connect corporate clients and cybersecurity specialists to Astra. In the coming days, subscribers of the Plus, Pro, Business, and Enterprise plans will gain access. Notably, the standard version of Astra will be included in the ChatGPT Plus subscription for $20 per month, making it even more attractive. The extended version, Astra Pro, will be available only on more expensive plans.
It is worth noting that the company has not yet officially called Astra artificial general intelligence. OpenAI President Greg Brockman called AGI a "gray, blurry thing," but allowed that over time this very model might be called "the point where AGI arrived." However, reasons for skepticism remain: the impressive result on the ARC-AGI-3 test was largely achieved through auxiliary features — an agent with memory and access to tools — which does not allow separating the pure capabilities of the model itself from the system built around it.
Creativity, for now, remains with humans. But Astra is an important step toward ChatGPT becoming not just a prompt assistant, but a tool that does the work for you. For a market where the speed of information processing and automation are becoming key factors, this event is certainly significant. The only question is how quickly we can adapt to new realities where AI takes on not only routine but also intellectual tasks.
My verdict: this is not just another update, but a paradigm shift. Investors and analysts should closely watch how the adoption of agentic AI will affect labor productivity and, consequently, macroeconomic indicators — this factor could become a new driver for the technology sector.