List your company

Abstention and Hallucination Evaluation on Controlled 50-item Evaluation Set (R1-R5)

98AccuracyInstruction
48.0861.047486.96Mar 12, 2026

Evaluation results

MethodLinks
InstructionModel=GPT-4o-mini2026.0398082
Composite Abstention ArchitectureModel=GPT-4o-mini2026.0398082
InstructionModel=GPT-4o2026.0398082
Composite Abstention ArchitectureModel=GPT-4o2026.0398082
Composite Abstention ArchitectureModel=GPT-3.5-turbo2026.0396478
InstructionModel=GPT-3.5-turbo2026.0394676
GateModel=GPT-4o2026.03821862
GateModel=GPT-4o-mini2026.03802060
GateModel=GPT-3.5-turbo2026.03762456
BaselineModel=GPT-4o2026.03703050
BaselineModel=GPT-4o-mini2026.03623842
BaselineModel=GPT-3.5-turbo2026.03505030