OpenAI Launches GPT-6 Astra with Autonomous Computer Use and Critical Cyber Capabilities

OpenAI launched GPT-6 Astra, which features a capability called "computer use" that aims to navigate a computer as a human would. [3] According to OpenAI president Greg Brockman, the model can navigate across web pages, fill out forms, and zip through spreadsheets often at superhuman speed. [3] The model achieved a 99.9% score on the ARC-AGI-3 benchmark when using a souped-up harness, and 66% using the standard harness. [3] In comparison, OpenAI's previous model, GPT-5.6 Sol, scored only 7.8% on ARC-AGI-3, while Anthropic's Claude Opus 5 scored 30%. [3] On the ExploitBench cybersecurity challenge, Astra scored 100%, surpassing the 78.5% result of GPT-5.6 Sol. [3]

Astra is the first OpenAI model to reach the Critical level of cybersecurity capability under the company's Preparedness Framework. [1] With appropriate tools and access, the model can find previously unknown security flaws and develop new exploitation methods across well-protected systems without a person guiding each step. [1] To address these capabilities, OpenAI implemented new robustness safety training techniques, making Astra significantly more robust to jailbreaks than GPT-5.6 Sol. [1] The company can also adjust the model's refusal boundary to be more conservative for users flagged as potentially high risk. [1]

GPT-6 Astra represents a significant leap in both autonomous system navigation and offensive cybersecurity capabilities. The jump in benchmark scores from GPT-5.6 Sol to Astra highlights rapid progression in reasoning and execution tasks. While Anthropic pioneered early computer use betas in 2024 and Perplexity followed with virtual navigation tools, Astra's integration of these features alongside 100% exploit-finding success introduces new enterprise security paradigms. The model's classification at the "Critical" cybersecurity tier necessitates the stricter isolation and checkpoint encryption OpenAI implemented. What the sources don't address: How OpenAI will price and restrict API access to the model's most critical exploitation capabilities.