OpenAI has launched GPT-6 Astra, a new artificial intelligence model engineered to tackle computer use, software development, cybersecurity, scientific research, and other professional tasks. Central to this release is the system’s ability to operate computers directly, navigating across spreadsheets and web pages, filling in digital forms, creating documents and presentations, and building websites before checking whether their features work as intended. OpenAI claims Astra completes these computer-based tasks faster than GPT-5.6 Sol while achieving higher performance. In simulations using the offline portion of OSWorld 2.0, the model recorded higher computer-use performance in about 47% less time per task.
OpenAI president Greg Brockman described this direct computer interaction as one of the primary advances in the model. Speaking on the new capabilities, he explained, “Computer use is a particularly important part of what’s new; the model can zip through spreadsheets, fill out forms, and navigate across web pages often at superhuman speed.”
Brockman also made a much broader claim regarding Astra’s place in the development of artificial intelligence. “If we fast forward a couple years, and we look back and say when was it really that AGI was created, I think it’s going to be about this time, and I think it might be about this model,” he stated. While the observation represents his personal assessment rather than evidence that artificial general intelligence has been achieved, it shows how strongly OpenAI is framing the release internally.
Astra takes on longer software and workplace tasks
Astra is designed to handle workflows involving several steps across different tools. In everyday workplace settings, OpenAI lists document creation, spreadsheet work, presentations, website development, and software testing among the tasks it can perform. Software development represents another area where the company reports gains. On DeepSWE v1.1, a benchmark covering software engineering tasks, Astra outperformed GPT-5.6 Sol while carrying an estimated API cost per task that was about 57% lower when comparing each model at its highest-scoring setting.
Early testing conducted by Canva gives an indication of the workload OpenAI expects Astra to handle. According to Canva, the model navigated a codebase containing more than 80 million lines, wrote and analysed more than 1,000 data queries, and drew on more than 21 internal knowledge sources to recommend improvements. The example covers a substantially larger working environment than a simple coding prompt, combining code, data queries, and information drawn from several internal sources. These results were reported by Canva as part of early customer testing.
Mathematics and cybersecurity capabilities expand
OpenAI is also applying Astra to mathematics and scientific research. An internal version of the model contributed to ten advances in mathematics and theoretical computer science, with proofs formalised in Lean, whilst Astra also scored 98% on FrontierMath Tier 4.
Cybersecurity is another area where OpenAI reports increased capability, though this expanded strength has prompted additional caution. Astra meets the Critical threshold for cybersecurity under the company’s Preparedness Framework, a classification that has prompted additional safeguards against misuse. OpenAI says the same capabilities can help security teams find and patch weaknesses. It plans to expand access for defensive cybersecurity work through OpenAI Daybreak, with less restrictive safeguards expected in the coming weeks for vulnerability validation, malware analysis, and detection engineering.
GPT-6 Astra is initially rolling out to a limited group of organisations. OpenAI plans to make it available over the coming days to ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS.
