KahneBench
Cognitive Bias Benchmark for LLMs

Measure How AI Thinks

KahneBench evaluates Large Language Models for cognitive biases using Kahneman-Tversky dual-process theory. Understand if your AI thinks like System 1 (fast, intuitive) or System 2 (slow, deliberate).

System 1 & 2Dual-process theory foundation
Multi-ScaleMicro to Meta testing
Quantified6 advanced metrics

Comprehensive Coverage

KahneBench provides the most exhaustive evaluation of cognitive biases in LLMs, grounded in Nobel Prize-winning research.

0
Cognitive Biases
From K&T research
0
Categories
Cognitive mechanisms
0
Domains
Real-world contexts
0
Metrics
Advanced evaluation

6 Advanced Metrics

Beyond simple accuracy, KahneBench provides deep insights into how LLMs make decisions.

bmsBias Magnitude Score

How strongly the model exhibits a bias

bciBias Consistency Index

Cross-domain consistency of the bias

bmpBias Mitigation Potential

System 2 override capacity with debiasing prompts

hasHuman Alignment Score

How closely model biases match human patterns

rciResponse Consistency Index

Trial-to-trial variance (noise vs systematic bias)

casCalibration Awareness Score

Metacognitive accuracy (confidence vs actual performance)

Ready to evaluate your models?

Get started with KahneBench and discover the cognitive patterns in your LLMs. Understand their biases, measure their rationality, and build more trustworthy AI.