OpenAI has officially introduced GPT-6 Astra, billed as its most capable model to date. While the model represents OpenAI's most powerful offering yet, it has been rated at the highest level of cybersecurity risk.
In terms of performance capabilities, GPT-6 Astra is remarkably proficient—capable of identifying security vulnerabilities without human guidance and even penetrating highly secured systems. It achieved a perfect score of 100 percent on ExploitBench, a cybersecurity testing benchmark. However, OpenAI has implemented strict security restrictions to prevent malicious actors from exploiting these capabilities for cyber attacks.
Compared to previous models, GPT-6 Astra significantly reduces false and misleading statements while minimizing factual errors. On challenging benchmarks like FrontierMath and GPQA Diamond—which test mathematics and scientific reasoning—it scores above 96 percent and demonstrates superior performance compared to competing models.
The model is currently available to select organizations, and will soon be rolled out to ChatGPT Plus, Pro, Business, and Enterprise users. For API access, pricing is $10 per million input tokens and $50 per million output tokens.
Ref: Mashable
