Cursor releases CursorBench 3.1 evaluation framework with standardized coding agent performance metrics

According to Cursor's evals site, CursorBench 3.1 establishes standardized performance benchmarks for AI coding agents. The framework ranks models including Fable 5, Opus 4.8, and GPT-5.5 across scoring percentage, cost per task, token usage, and steps per task, providing the first major comparative benchmark tool for agent coding performance.