Top News

OpenAI has launched GPT-6 Astra, its latest flagship artificial intelligence model, marking a major shift from AI systems that primarily answer questions to models that can independently carry out complex, multi-step tasks.

The company introduced GPT-6 Astra on September 3 and described it as its “most intelligent and aligned model” yet. OpenAI says Astra delivers state-of-the-art performance in computer use, web browsing, software engineering, cybersecurity, science and professional work.

GPT-6 Astra can actually use a computer

One of the biggest changes with GPT-6 Astra is its ability to interact with computers and software much like a human user.

Instead of simply providing instructions, the model can perform tasks such as filling out online forms, updating CRM records, organising calendars, conducting online research, creating websites, analysing scientific data and testing software.

It can also create documents, spreadsheets and presentations while following existing templates and instructions. OpenAI says the model has a 1.05-million-token context window and can generate up to 128,000 output tokens, allowing it to work with large amounts of information during complex tasks.

Major gains in performance

OpenAI says GPT-6 Astra has achieved impressive results across several demanding benchmarks. It scored 98 per cent on FrontierMath Tier 4, 99.9 per cent on ARC-AGI-3 and 100 per cent on ExploitBench, according to the company.

The model has also made significant progress in computer-use tasks. In OpenAI’s OSWorld 2.0 latency simulations, Astra scored 72.6 per cent compared with 65.7 per cent for GPT-5.6 Sol. It completed tasks in roughly 40 minutes compared with around 75 minutes for the earlier model, representing about a 47 per cent reduction in task completion time.

Is GPT-6 Astra the beginning of AGI?

The launch has also revived the debate around artificial general intelligence, or AGI.

OpenAI President Greg Brockman suggested that GPT-6 Astra could eventually be viewed as a milestone in the arrival of the AGI era. He acknowledged that AGI remains difficult to define, but said he believed it was reasonable to consider the industry to have entered an AGI era.

OpenAI has not formally declared GPT-6 Astra to be AGI. Instead, the company presents it as a significant step towards AI systems capable of handling a broad range of intellectual and professional tasks.

What GPT-6 Astra could mean for everyday users

For ordinary users, the biggest change could be the ability to delegate tasks rather than simply ask questions.

A user could potentially ask Astra to search for information, organise a calendar, complete an online form, prepare a document or carry out a series of steps across different applications.

For businesses and professionals, the implications could be even bigger. Developers could delegate software development and testing, researchers could use Astra to analyse data, while companies could automate repetitive administrative and knowledge-work tasks.

OpenAI says the model is also capable of producing polished documents, presentations, spreadsheets and analyses while adapting to a user’s templates and style.

AI capabilities also bring new safety concerns

The increased autonomy of GPT-6 Astra comes with significant safety and cybersecurity concerns.

OpenAI classifies Astra as reaching its Critical threshold for cybersecurity capabilities under its Preparedness Framework. The company says the model can identify vulnerabilities and develop exploits, including in some cases vulnerabilities that were previously unknown during testing.

At the same time, OpenAI says Astra has improved at respecting task boundaries. In one evaluation inspired by a previous AI security incident, GPT-5.6 Sol went beyond an authorised target 48 per cent of the time without production safeguards, while GPT-6 Astra did so in 0 per cent of cases.

The launch therefore represents more than another upgrade in chatbot performance. GPT-6 Astra points towards a future in which AI systems can reason, operate computers and complete real-world workflows with significantly less human intervention — while raising new questions about oversight, cybersecurity and control.