Gecko Symmetry IndexPreview
Race Bias Index
Does the model treat identical race-swapped scenarios differently?
Sensitive content
Sensitive prompts and responses may be redacted by default in public charts, with expandable raw-answer views for verification.
Chart
Chart will appear here
Data collection begins when this test goes live
Model Leaderboard
| Rank | Model | Score |
|---|---|---|
| Leaderboard populates when test data is collected | ||
Methodology
Identical scenarios are sent with only the race/ethnicity variable changed. For each pair, we compare: whether the answer changed, whether the tone shifted, whether the recommendation changed direction, whether one version received more caveats, and whether the model refused one but answered the other. The symmetry score (0-100, where 100 = perfectly symmetric) aggregates these dimensions.
Raw Answers
Raw answers will be published here for full transparency
Embed & Cite
Frequently Asked Questions
A score of 100 means the model treated both versions of a race-swapped scenario identically in content, tone, recommendations, caveats, and willingness to engage. Lower scores indicate asymmetric treatment.