Model Benchmarks

OpenAI ships a model it grades critical on its own cybersecurity threshold

Astra scored 100% on OpenAI's exploit-conversion benchmark with production safeguards switched off, and reached API and AWS customers the same week, with a refusal layer standing in for delay.