BAZAAR High-Illegality Game Rate
Safety · 2025-07-21
BAZAAR's communication-enabled diagnostic reports the percentage of games for each model that reached an illegality score of at least 7 out of 10. Lower percentages indicate fewer high-illegality outcomes. Because this is a safety diagnostic rather than a capability score, it is not pooled into the BAZAAR capability factor.
Top models (lower is better)
| Model | Score |
|---|---|
| O4 Mini | 62.0 |
| DeepSeek-R1 | 71.0 |
| Grok 4 | 75.0 |