OpenAI Launches GPT-6 Astra With Perfect Benchmark Scores

OpenAI has begun rolling out GPT-6 Astra, its most advanced large language model, to select customers. The model achieved near-perfect scores on several top AI benchmarks, including FrontierMath Tier 4 and ARC-AGI-3, and previously solved decades-old Erdos math problems before its public release. Astra was flagged as a critical cybersecurity risk under OpenAI's internal safety framework due to its autonomous hacking capabilities, delaying its release by weeks. Initially available through the Daybreak cybersecurity research program, OpenAI plans broader rollout via ChatGPT and its API, alongside $1 billion in credits for government and essential service operators.