© ROOT-NATION.com - Use of content is permitted with a backlink.
A little less than two months after OpenAI released its previous major lineup of GPT-5.6 Sol, Terra, and Luna models to the public, the company is back with a completely new development called GPT-6 Astra. According to the developer itself, this is the most intelligent and coherent model on the planet. Given OpenAI’s decision to slow down the development of advanced systems in August 2026 following an incident in which one of its algorithms crashed the Hugging Face AI platform, this release may turn out to be the company’s last major launch for the foreseeable future.
According to OpenAI, GPT-6 Astra demonstrates exceptional performance in autonomous PC management and web browsing, and excels at tasks in the fields of software development, cybersecurity, science, and everyday professional activities. In a striking demo video accompanying the announcement, OpenAI showed how Astra easily handles a wide variety of tasks – from 3D modeling to creating presentations. At the same time, the model often performs several tasks from different fields simultaneously – for example, ordering food while writing code for a game. A key advantage of Astra is its ability to perform such multi-step workflows directly on your device, as well as both within and outside a web browser. OpenAI assures us that the model carries out these complex tasks with high visual accuracy and without deviating from the original instructions or query.
As with previous presentations, OpenAI has prepared a new set of test results to demonstrate Astra’s growing capabilities. Among them is a score of 98.6% on the ARC-AGI-3 test, which is considered one of the key industry benchmarks for evaluating an AI’s ability to solve unfamiliar problems. VentureBeat notes that this result comes with certain caveats. These stem from the fact that the systems taking this test differ in their configurations, and the variety of system architectures (such as the presence of persistent memory) can influence the final results. Other achievements appear far more indisputable: Astra scored 57.7% on the Terminal Bench 4.0 specialized coding test and 59.3% on the Agent’s Last Exam, a test of autonomous capabilities, significantly outperforming GPT-5.6 Sol. OpenAI included even more results in its announcement, but to summarize, Astra currently tops most performance rankings.
This raises the broader question of whether such an AI system can be used and deployed responsibly. For example, Astra possesses impressive skills in the field of cybersecurity. OpenAI reported that the model scored a perfect 100% on the ExploitBench test, which assesses algorithms’ ability to find and exploit vulnerabilities in software. This represents a significant leap forward compared to the 78.5% score achieved by GPT-5,6 Sol. In addition, OpenAI tested Astra on the SRE-Bench test, which measures an algorithm’s ability to reverse-engineer binary files without access to the source code. It turned out that Astra successfully completed 88.0% of the tasks on the first attempt and 99.2% within four attempts, while GPT-5,6 Sol’s results were only 55.9% and 68.7%, respectively.
Astra’s high level of alignment – which encompasses everything from strict adherence to specified templates to increased transparency in communication – is expected to raise the bar for security. However, OpenAI emphasizes that it has also further restricted the model so that it refuses to perform complex tasks related to cyberattacks. According to OpenAI, these enhanced safeguards will help Astra better resist attempts to circumvent security protocols while also improving the company’s own mechanisms for detecting misuse.
OpenAI notes that the rollout of GPT-6 Astra begins today for a limited group of organizations, and in the coming days, the model will be made available to all ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as via the OpenAI API and AWS platforms. For developers working with the OpenAI API, the pricing for Astra is set at $10 per 1 million input tokens and $50 per 1 million output tokens.
See also: A New Era of Risks: OpenAI Confirms Astra’s Super-Powered Cyber Capabilities