Agent Arena Benchmarks Show Anthropic Claude Leading in Code and Knowledge Work

Agent Arena, a benchmark platform tracking agentic AI model performance across real-world tasks, reveals Anthropic's Claude Fable 5.1 commanding leads in code generation and knowledge-work domains. The October 2026 leaderboard shows Claude Fable 5.1 (Max) achieving 17.87% net improvement in code tasks and 16.91% confirmed success in work-related assignments, both first place. OpenAI's GPT-6 Astra (Max), released September 3, ranks second in code at 16.92% but places fourth in chat applications.