OpenAI’s GPT-6 Astra Ranks Second in Agent Arena Debut

OpenAI’s GPT-6 Astra (Max) has entered Agent Arena at No. 2 overall, recording a 12.55% net improvement across 8,929 real-world agentic sessions. The result places it behind Anthropic’s Claude Fable 5.1 (Max), which posted a 14.51% net improvement on the leaderboard dated September 8, 2026. Agent Arena’s leaderboard measures models on tool orchestration for real-world agentic tasks, including task completion, tool reliability and steerability.