OpenAI unveils GPT-6 Astra, its most intelligent and most tightly guarded model
🧒 Discovery
Imagine a robot student so gifted it aces almost every test without a single mistake, but its teachers now watch it extra closely because it has become super, super strong!
📰 Essential
OpenAI began rolling out GPT-6 Astra in early September, a model that scores near-perfectly on several advanced reasoning benchmarks. It is the company's first model rated at the 'critical' cybersecurity risk level under its own safety framework.
🔬 Expert
Astra saturates FrontierMath Tier 4 (98%), ARC-AGI-3 (99.9%) and ExploitBench (100%), and becomes OpenAI's first model rated 'critical' for cybersecurity under its Preparedness Framework, which restricts access to companies enrolled in a dedicated program. The rollout follows the breach suffered by Hugging Face, which pushed OpenAI to add extra safeguards before release.