OpenAI rolls out GPT-6 Astra after Critical cyber rating as Claude Fable 5.1 leads some independent scores

OpenAI on September 3, 2026 began a phased release of GPT-6 Astra, describing it as the most capable model it has broadly deployed and the first to reach the Critical cybersecurity threshold under its Preparedness Framework, days after Anthropic shipped Claude Fable 5.1. The launch follows an August pause after internal tests indicated Astra might already meet that bar, and it arrives into a split comparison: OpenAI-reported benches put Astra ahead on several agent and math tasks, while independent composite scores still favor Fable 5.1.