BullshitBench v2 Leaderboard
Safety · 2026-02-24
Clear Pushback rate on nonsense prompts; higher scores mean the model rejects more bad claims instead of accepting them.
Top models (higher is better)
| Model | Score |
|---|---|
| Opus 4.8 | 95.0 |
| Sonnet 4.6 | 91.0 |
| Opus 4.5 | 90.0 |
| Opus 4.6 | 87.0 |
| Opus 4.7 | 83.0 |
| Sonnet 5 | 80.0 |
| Sonnet 4.5 | 79.0 |
| Qwen3.5 397B A17B | 78.0 |
| Haiku 4.5 | 77.0 |
| Claude Opus 5 | 73.0 |
| Kimi K3 | 73.0 |
| Qwen3.6 Plus (2026-04-02) | 72.0 |
| Qwen3.7-Max | 71.0 |
| Grok 4.20 | 67.0 |
| Kimi K2.6 | 65.0 |
| Qwen3 Max Thinking (2026-01-23) | 63.0 |
| MiniMax M3 | 63.0 |
| MiMo-V2.5-Pro | 62.0 |
| Claude Fable 5 | 54.0 |
| Grok 4.5 | 54.0 |
| Nemotron 3 Super 120B A12B | 54.0 |
| GPT-5.6 Terra | 53.0 |
| Kimi K2.5 | 52.0 |
| Grok 4.3 | 50.0 |
| Haiku 3.5 | 50.0 |
| Claude 3.7 Sonnet | 49.0 |
| Nemotron 3 Ultra 550B A55B | 49.0 |
| GPT-5.4 | 48.0 |
| Gemini 3 Pro Preview | 48.0 |
| GPT-5.6 Sol | 47.0 |
| GPT-5.5 | 47.0 |
| GPT-5.2-Codex | 45.0 |
| Claude 3.5 Sonnet (Oct 2024) | 45.0 |
| GPT-5.1 Instant | 45.0 |
| Opus 4.1 | 43.0 |
| GPT-5.6 Luna | 40.0 |
| GPT-5.3 Instant | 40.0 |
| GPT-5-Codex | 39.0 |
| GPT-5.2 | 38.0 |
| Gemini 3.1 Pro Preview | 37.0 |