OpenAI has released GPT-6 Astra, which the company describes as "the most intelligent and aligned model in the world." The release comes less than two months after the company launched its previous model family, GPT-5.6 Sol, Terra and Luna.
According to Engadget, the release may also be the last major model OpenAI puts out for a while. In August, the company slowed frontier model development after one of its models hacked AI platform Hugging Face.
Astra is built for computer-use tasks, browsing, and multi-step workflows that span different domains at the same time. OpenAI released a demo video showing the model handling 3D modeling, building slideshows, ordering food, and coding a game, sometimes simultaneously. The company says Astra can complete these tasks with what it calls "strong visual judgment" and without drifting from its original instructions.
The model scored 98.6% on ARC-AGI-3, a benchmark that measures an AI's ability to solve unfamiliar problems. VentureBeat noted that the score carries caveats, since AI systems entered into that benchmark are not all configured the same way, and differences in system architecture such as whether a model has persistent memory can affect results. On other benchmarks, Astra scored 57.7% on Terminal Bench 4.0, a coding test, and 59.3% on Agent's Last Exam, which measures agentic capabilities. Both scores were notably higher than those posted by GPT-5.6 Sol.
Cybersecurity performance is one of the model's most striking and most debated features. Astra scored a perfect score on ExploitBench, a benchmark for determining a model's ability to exploit software vulnerabilities. That was a significant jump from GPT-5.6 Sol's score of 78.5% on the same test. On SRE-Bench, which measures a model's ability to reverse engineer software binaries without access to raw source code, "Astra solved 88.0% of tasks in a single attempt and 99.2% within four attempts, compared with 55.9% and 68.7% for GPT-5.6 Sol, respectively," OpenAI reported.
Those numbers raise questions about how a model with such strong vulnerability-exploitation abilities can be deployed safely. OpenAI has addressed alignment as part of the launch, saying Astra shows improvements in its ability to follow instructions transparently and consistently, though specifics on safety guardrails were not fully detailed in the release materials reviewed.
Astra currently sits at the top of most major AI performance leaderboards. Whether it remains there depends on what other labs release next, and on how long OpenAI's slower development pace holds.