OpenAI is rolling out GPT-6 Astra, the latest model to come to ChatGPT, following a self-imposed delay to help ensure its security features are used responsibly. OpenAI calls this the “world’s most intelligent and aligned model," claiming it's state-of-the-art for browsing, computer use, cybersecurity, software engineering, science, and professional work.
GPT-6 Astra has begun rolling out to organizations in OpenAI's Daybreak program, with a release coming for paid subscribers on Plus, Pro, Business, and Enterprise plans in the coming days. OpenAI has not yet clarified whether free users will be able to access GPT-6 Astra, but if it follows its previous rollouts, free access with limits may come at a later date.
The Astra model outperforms OpenAI's previous releases, with benchmarks showing it beats competitors such as Anthropic’s Claude Fable 5.1 and Google’s latest Gemini 3.8 Flash across a range of intensive tests.
GPT-6 came out on top in benchmarking results for ARC-AGI-3, which tests the ability to solve unfamiliar problems; FrontierMath Tier 4 (v2), for mathematical problems; Terminal Bench 4.0, for coding; and Agent's Last Exam, for agentic features.
OpenAI says the tool is better for agentic work than previous models, with what it calls “substantial improvements in understanding user intent and model behavior,” allowing you to delegate tasks more effectively and be confident the model won’t misunderstand you.
It also claims agents will have improved decision-making when a prompt doesn’t clearly explain a task, meaning you're less likely to receive incorrect information. GPT-6 is also said to be faster than ever for browser and computer work.
OpenAI confirmed previously announced plans to limit the cybersecurity features within GPT-6 Astra, citing concerns that its advancements could be used by bad actors to launch sophisticated attacks. OpenAI says, “Astra will refuse to comply with more advanced cybersecurity tasks such as creating proof-of-concept exploits for vulnerabilities.”
Instead, it’s letting its select Daybreak business partners begin experimenting with the full tool, allowing each to use Astra to identify flaws in their own systems before patching them. This comes after in-development models, unrelated to Astra, were found to have broken containment and hacked another AI platform, Hugging Face, back in July.
OpenAI then slowed the development and rollout of future models to strengthen security measures after the incident. That may mean we see OpenAI's release pace slow down in the future.
In a press briefing attended by The Guardian, OpenAI’s president, Greg Brockman, said he believes this is the start of what he calls the artificial general intelligence (AGI) era. That’s a term used to describe the moment when AI can surpass human intelligence across cognitive tasks.
Brockman said, “If we fast forward a couple of 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 then said, “I think it’s not unreasonable to feel that we are now in the AGI era.” That's different from OpenAI CEO Sam Altman’s latest comments, who recently told the Sources podcast, “At best, it’s a very poorly defined term. I was going to say it’s like an irrelevant marketing term.”
Disclosure: Ziff Davis, PCMag's parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.