OpenAI Launches GPT-6 Astra: The Flagship Model Designed to Complete Tasks, Not Just Answer Questions

OpenAI officially launched GPT-6 Astra on Thursday, September 3, introducing what the company describes as the world’s most intelligent and aligned AI model. Developed under the leadership of CEO Sam Altman, the new model marks a significant departure from earlier conversational AI systems — shifting the focus from answering questions to independently completing complex, real-world tasks.

What Is GPT-6 Astra and What Can It Do?

Unlike previous AI models that responded to prompts with text-based answers, GPT-6 Astra is built to interact with computers much like a human user would. OpenAI says the model is state-of-the-art across computer use, web browsing, software engineering, cybersecurity, scientific research, and professional work.

In practical terms, Astra can:

• Browse websites and gather information autonomously
• Fill out forms and update CRM records
• Organise calendars and schedule appointments
• Write, test, and debug software code
• Analyse scientific data and conduct research
• Create documents, spreadsheets, and presentations using a user’s own templates
• Build and iterate on websites

The model features a 1.05-million-token context window, enabling it to process and work with enormous volumes of information within a single task. It can also generate up to 128,000 output tokens, making it suitable for large-scale, multi-step workflows. OpenAI describes Astra as the product of years of research combining advances in pre-training, reinforcement learning, and alignment.

Benchmark Performance: How Does GPT-6 Astra Compare?

OpenAI has published benchmark results that position Astra well ahead of its predecessors across several demanding evaluations.

On the OSWorld 2.0 benchmark, Astra completed tasks in roughly 40 minutes compared to 75 minutes for GPT-5.6 Sol — a reduction of approximately 47 per cent in time per task. Legal technology company Legora reported that Astra improved performance on a financial-statement workflow by around 40 per cent and successfully identified all four deliberately planted errors across 41 documents.

Is GPT-6 Astra the Beginning of AGI?

Perhaps the most significant aspect of the launch was not the benchmarks but the language used by OpenAI’s leadership. During a press briefing ahead of the announcement, OpenAI President Greg Brockman suggested that GPT-6 Astra could eventually be viewed as a milestone in the arrival of artificial general intelligence (AGI) — broadly understood as AI capable of learning, reasoning, and performing a wide range of intellectual tasks at a human level.

Brockman acknowledged that AGI remains a “grey, fuzzy thing,” but said there is a strong case that the technology has now entered that era. “I think it’s not unreasonable to feel that we are now in the AGI era,” he said. When pressed on whether OpenAI was formally declaring AGI achieved, Brockman clarified that the term was no longer tied to the company’s agreement with Microsoft. He described AGI as a “mission concept or spiritual concept.” He reportedly closed the briefing with the statement: “Welcome to the AGI era.”

What GPT-6 Astra Means for Everyday Users and Professionals

For everyday users, the most immediate change is in how tasks can be delegated. Rather than asking an AI to suggest how to book a doctor’s appointment or compare restaurants, users can simply instruct Astra to perform those tasks directly. The model can navigate websites, operate software, and complete multiple sequential steps without requiring instructions at each stage.

For professionals, the implications are broader:

• Developers can delegate portions of software development and testing
• Researchers can ask Astra to gather, organise, and analyse information independently
• Businesses can automate repetitive administrative processes
• Creative teams can use Astra to build and iterate on products more rapidly

OpenAI says this shift could allow AI to handle time-consuming tasks faster than people typically can.

Safety, Cybersecurity, and the Risks of More Capable AI Agents

GPT-6 Astra’s capabilities also raise significant concerns. OpenAI’s own safety evaluations classify Astra as reaching its Critical cybersecurity capability threshold. The company confirmed the model can identify software vulnerabilities and develop exploits, including previously unknown ones in some testing environments.

This presents a clear trade-off: the same capabilities that help developers identify security flaws could potentially be exploited by malicious actors. OpenAI has also pointed to improvements in task boundary compliance as evidence of stronger alignment. In a test modelled on a previous AI security incident, GPT-5.6 Sol exceeded an authorised target 48 per cent of the time when production safeguards were absent, while Astra did so in 0 per cent of cases.

OpenAI maintains that Astra is its most aligned model to date, though questions around human oversight, control mechanisms, and misuse potential remain central to ongoing debate about increasingly autonomous AI systems. The launch of GPT-6 Astra represents a meaningful shift in what AI systems are expected to do — moving well beyond conversation and into independent, real-world action.