Agent Evaluations

Evaluate Claude agents credibly. Learn task-based eval suites (SWE-bench, METR, AgentBench, Anthropic-published agent evals), success metrics (pass rate, partial credit, cost-per-success), regression suites for production agents, the offline-vs-online eval split, the link to production monitoring (success rate per task class, drift detection), and the failure modes of weak eval design.

Start Topic → View All Lessons
6
Lessons
📋
Templates
✅
Practitioner-Ready
100%
Free

Lessons in This Topic

Work through these 6 lessons in order, or jump to whichever is most relevant.

📊

Agent Evaluations Overview

Advanced
📊

Task-Based Eval Suites

Advanced
📊

Success Metrics

Advanced
📊

Regression Suites

Advanced
📊

Offline vs Online Eval

Advanced
📊

Eval-to-Production Monitoring

Advanced