Sensitive Content Reasoning on SenBen 2,000 frames (test)
0.652RSBGemini 3 Pro
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini 3 ProReasoning mode=low reas., Zero-shot inference=true2026.04 | 0.652 | 64.7 | |
| Gemini 3 FlashReasoning mode=low reas., Zero-shot inference=true2026.04 | 0.593 | 58.3 | |
| Q2L-aggParams=241M2026.04 | 0.449 | 43.1 | |
| Q2L-balParams=241M2026.04 | 0.413 | 42.8 | |
| Claude Opus 4.6Zero-shot inference=true2026.04 | 0.327 | 40.4 | |
| GPT-5.2Reasoning mode=med. reas., Zero-shot inference=true2026.04 | 0.319 | 36.2 | |
| GLM-4.6VParams=10.3B, Reasoning mode=reas., Zero-shot inference=true2026.04 | 0.291 | 36.4 | |
| Qwen3-VL-8BParams=8.3B, Zero-shot inference=true2026.04 | 0.286 | 34 | |
| GPT-5-miniReasoning mode=med. reas., Zero-shot inference=true2026.04 | 0.285 | 33 | |
| Claude Sonnet 4.6Zero-shot inference=true2026.04 | 0.277 | 33.9 | |
| GPT-5.2Zero-shot inference=true2026.04 | 0.247 | 30.4 |