General Visual Question Answering on HallusionBench
69.9Pass@1Gemini-3-Pro High
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3-Pro High2026.06 | 69.9 | |
| Seed2.0 Pro2026.06 | 68 | |
| GPT-5.2 High2026.06 | 67.7 | |
| Seed2.0 Lite2026.06 | 66 | |
| Claude-Opus-4.52026.06 | 65.3 | |
| Seed2.0 Mini2026.06 | 65.1 | |
| Seed1.82026.06 | 63.9 | |
| Gemini 1.5 Prothinking=true, decoding=greedy2025.05 | 63.7 | |
| Seed 1.5-VLthinking=true, decoding=greedy2025.05 | 60.3 | |
| Seed 1.5-VLthinking=false, decoding=greedy2025.05 | 60 | |
| Claude 3.7 Sonnetthinking=true, decoding=sampling2025.05 | 58.3 | |
| GPT-4othinking=false, decoding=greedy2025.05 | 56.2 | |
| OpenAI o1thinking=true, decoding=greedy2025.05 | 55.6 | |
| Qwen 2.5-VL 72Bthinking=false, decoding=greedy2025.05 | 55.2 |