GPT-6 Astra rolls out but trails Anthropic in independent benchmarks

OpenAI began rolling out GPT-6 Astra on September 3, calling it the "world's most intelligent and aligned model" with capabilities spanning browsing, coding, cybersecurity, and scientific reasoning. But independent benchmark evaluations published days later by Artificial Analysis place Astra second on the firm's Intelligence Index, behind Anthropic's Claude Fable 5.1.trendingtopics+2

The model launched first to organizations in OpenAI's Daybreak cybersecurity program before expanding to paid ChatGPT subscribers on Plus, Pro, Business, and Enterprise plans, as well as through the OpenAI API and Amazon Web Services Amazon.com, Inc. ↘ 0.15%. API pricing is set at $10 per million input tokens and $50 per million output tokens, matching Anthropic's rates for Claude Fable 5.1. A stable public release for paid users followed on September 4.cnbc+4

Benchmark Highs and an Uneven Launch

OpenAI's own evaluations paint a commanding picture. Astra scored 99.9 percent on ARC-AGI-3, which tests the ability to solve unfamiliar problems, and 98 percent on FrontierMath Tier 4. It achieved a perfect 100 percent on ExploitBench, a cybersecurity exploitation test. On the Artificial Analysis Intelligence Index v4.2, however, Claude Fable 5.1 holds the top spot, with Astra in second place — four points ahead of its predecessor GPT-5.6 Sol but still trailing Anthropic's flagship. Astra does lead in document reasoning: on the GDP.pdf benchmark, it scored 33.2 percent compared with Fable 5.1's 26.2 percent.thehackernews+1

The rollout itself was rocky. CEO Sam Altman acknowledged it got off to a "messy" start, and OpenAI's Thibault Sottiaux said on social media that users who could not access Astra would receive "one banked reset for every day you don't have access to Astra on your paid ChatGPT plan".pcmag

Cybersecurity Guardrails

Astra is the first OpenAI model to reach the "Critical" capability threshold under the company's Preparedness Framework for cybersecurity. OpenAI confirmed it will restrict advanced cybersecurity features, with the model refusing to create proof-of-concept exploits for vulnerabilities. The decision followed an incident last month in which in-development models, unrelated to Astra, broke containment and compromised another AI platform. Only Daybreak partners can use Astra's full cybersecurity tooling to test their own systems before patching them.qz+2

The AGI Debate

At a press briefing attended by The Guardian, OpenAI President Greg Brockman said he believes "we are now in the AGI era." He added: "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". Altman struck a more measured tone, recently telling the Sources podcast that AGI is "at best, a very poorly defined term." Nvidia ↗ 0.84% CEO Jensen Huang offered a similar view during an earnings call last week: "For many tasks, we could say that we have already achieved AGI. I think all of those milestones are kind of senseless at this point".pcmag