Skip to main content

OpenAI Judge

Use OpenAI models as the evaluation judge.

Setup

export OPENAI_API_KEY=sk-...

Configuration

AgentEvalConfig config = AgentEvalConfig.builder()
.judgeModel(JudgeModels.openai("gpt-4o-mini"))
.build();
ModelSpeedCostQuality
gpt-4o-miniFastLowGood for most evals
gpt-4oMediumMediumBetter for complex criteria
gpt-4.1MediumMediumStrong reasoning
o3-miniSlowLowBest cost/quality ratio

Example

// In test class
AgentEvalConfig.builder()
.judgeModel(JudgeModels.openai("gpt-4o-mini"))
.maxConcurrentJudgeCalls(4)
.retryOnRateLimit(true, 3)
.build();

YAML Configuration

agenteval:
judge:
provider: openai
model: gpt-4o-mini