Method predicts AI agent benchmark scores from fewer test runs

Original: Task-Efficient Agent Evaluation Using Cached Traces

Source: arxiv.org ↗