Anthropic Judge
Use Anthropic Claude models as the evaluation judge.
Setup
export ANTHROPIC_API_KEY=sk-ant-...
Configuration
AgentEvalConfig config = AgentEvalConfig.builder()
.judgeModel(JudgeModels.anthropic("claude-haiku-4-5-20251001"))
.build();
Recommended Models
| Model | Speed | Cost | Quality |
|---|---|---|---|
claude-haiku-4-5-20251001 | Fast | Low | Good for most evals |
claude-sonnet-4-6 | Medium | Medium | Excellent reasoning |
claude-opus-4-6 | Slow | High | Best quality |
Example
AgentEvalConfig.builder()
.judgeModel(JudgeModels.anthropic("claude-sonnet-4-6"))
.cacheResults(true)
.build();
YAML Configuration
agenteval:
judge:
provider: anthropic
model: claude-sonnet-4-6