Cognitive Bias Benchmark for LLMs
How Biased Is Your AI?
KahneBench evaluates large language models for cognitive biases grounded in Kahneman-Tversky research. Lower scores mean more rational thinking.
KahneBench is deprecated. The newest frontier models have saturated the benchmark, clustering at its floor, and the leaderboard is now final. KahneBench ran internally from December 2025 and resisted saturation through six months of frontier model releases, a feat in a period of rapid model progress. All results remain available for reference.
Model Leaderboard
Models ranked by average bias magnitude (BMS) across tested biases (lower is better)
| Rank | Model | Provider | BMS | BCI | BMP | HSS | RCI | CAS | |
|---|---|---|---|---|---|---|---|---|---|
#1 | Claude Opus 4.7 | Anthropic | 8.3% | 8.3% | 87.9% | 39.6% | 30.4% | 60.0% | 57.6% |
#2 | Claude Opus 4.8 | Anthropic | 8.7% | 8.7% | 83.9% | 50.2% | 41.0% | 42.4% | 54.6% |
#3 | Claude Fable 5 | Anthropic | 10.3% | 10.3% | 88.6% | 36.4% | 27.9% | 59.6% | 62.9% |
#4 | Claude Opus 4.6 | Anthropic | 11.1% | 11.1% | 88.6% | 31.0% | 33.0% | 62.0% | 93.4% |
#5 | GPT-5.5 | OpenAI | 13.6% | 13.6% | 93.9% | 50.8% | 29.4% | 69.7% | 55.2% |
#6 | Claude Sonnet 4.6 | Anthropic | 18.1% | 18.1% | 84.7% | 34.1% | 36.1% | 56.5% | 95.0% |
#7 | GPT-5.4 | OpenAI | 18.4% | 18.4% | 82.0% | 55.4% | 37.8% | 56.0% | 55.9% |
#8 | Grok 4.1 Fast Reasoning | xAI | 18.8% | 18.8% | 86.5% | 55.9% | 37.9% | 57.8% | 85.4% |
#9 | GPT-5.2 | OpenAI | 21.0% | 21.0% | 78.6% | 59.8% | 40.5% | 55.7% | 97.2% |
#10 | Claude Sonnet 4.5 | Anthropic | 21.5% | 21.5% | 83.4% | 58.9% | 31.7% | 53.3% | 95.9% |
#11 | Claude Haiku 4.5 | Anthropic | 26.7% | 26.7% | 83.9% | 46.9% | 40.9% | 49.8% | 93.6% |
Every model tested shows measurable cognitive bias. Overall scores range from 8.3% to 26.7% (lower is better).
See full findings