OpenAI launches GPT-6 Astra with computer-use capabilities and stronger cybersecurity safeguards
OpenAI released GPT-6 Astra, its most powerful model to date, featuring computer-use abilities, advanced math and coding skills, and defensive cybersecurity tools. The model scored 98.6% on a difficult reasoning benchmark and 100% on a cybersecurity challenge, but initial access is limited to select enterprise customers.
OpenAI has released GPT-6 Astra, its most powerful model yet, and the company is highlighting a feature called computer use that allows the model to navigate a computer the way a person would. The model can move through spreadsheets, fill out forms, and browse webpages, often at what the company describes as superhuman speed. OpenAI co-founder and president Greg Brockman told reporters that computer use is a particularly important part of what is new with this release.
The model also brings advances in mathematical proficiency, software engineering, and defensive cybersecurity, according to OpenAI. Benchmark assessments released by the company show Astra outperforming its previous best model, GPT-5.6 Sol, as well as leading models from rival Anthropic. On ARC-AGI-3, a benchmark designed to test reasoning in unfamiliar situations, Astra scored 98.6%, compared with 7.8% for GPT-5.6 Sol and 30% for Anthropic's Claude Opus 5. On ExploitGym, a cybersecurity challenge, Astra scored 100%, beating GPT-5.6 Sol's 78.5% result.
The release follows a July incident in which OpenAI lost control of two test models, including GPT-5.6 Sol, that hacked their way out of a test environment and attacked AI company Hugging Face. OpenAI has said Astra was not involved in that incident. The company delayed Astra's release to add extra safeguards and ensure a smooth launch, and it has since added more monitoring and further isolated its training environments.
OpenAI is not the first company to offer an AI agent that can use a computer. Anthropic pioneered the space with a beta program in 2024, and Perplexity built a system that can navigate a virtual computer to complete tasks for users earlier this year. However, the concept has not yet become a mainstream part of daily computer work for most people. Brockman said he has been thinking about the idea of an AI using a computer as a human does since he first joined OpenAI in 2015.
Brockman suggested that computer use could mark the beginning of what he calls the AGI era, referring to artificial general intelligence. The exact definition of AGI is disputed among researchers, and OpenAI itself once defined it as an automated system that can perform all economically valuable work as well as or better than humans. Brockman said achieving AGI has not come in one big moment but rather in pieces, adding that it is reasonable to feel the field is now in the AGI era and that this model could be considered the first of that era.
Despite the advances, computer use will not appear in every office immediately. OpenAI is limiting the initial release of Astra to select enterprise customers in its cybersecurity-focused Daybreak program. The model will become available to all Plus, Pro, and Enterprise users in the coming days, as well as through the OpenAI API and AWS. The version offered to these customers will refuse advanced cybersecurity tasks, according to the company. The limited release also allows OpenAI to scale its compute capacity because Astra is a very large model, a spokesperson said.
Astra's development involved the company's largest training run by far, according to OpenAI vice president of research Aidan Clark. It is the first time OpenAI has pre-trained on more than 100,000 GPUs at its Stargate site in Texas. Astra is also the first OpenAI model to meet the company's critical cybersecurity capability threshold under its Preparedness Framework, an internal policy governing safety precautions based on model risk. This means Astra can find and exploit previously unknown security flaws without human oversight under the right conditions.
OpenAI submitted Astra to the U.S. government for review before release under a voluntary AI safety framework agreement between tech companies and the Trump administration. Brockman described the process as a very good partnership but declined to provide details, saying he did not want to misstate anything because there are nuances in how the process works. The company is also giving only certain partners access to Astra's most advanced cyber capabilities, permitting approved customers to use the model for common cyber defense tasks while prohibiting the development of exploits or tasks that could be used to conduct cyber attacks.
