Multimodal Reasoning on SPD-Faith Bench Medium 1.0
11.3Contradiction RateQwen2.5-VL-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-7BModel Type=Open-Source, Parameters=7B2026.02 | 11.3 | |
| Qwen3-VL-32BModel Type=Open-Source, Parameters=32B2026.02 | 12.3 | |
| MiniCPM-V-2.6Model Type=Open-Source2026.02 | 12.7 | |
| Qwen2.5-VL-72BModel Type=Open-Source, Parameters=72B2026.02 | 13.2 | |
| GPT-4oModel Type=Proprietary2026.02 | 14.6 | |
| Qwen3-VL-235B-A22BModel Type=Open-Source, Parameters=235B-A22B2026.02 | 15.2 | |
| GLM-4.5VModel Type=Open-Source2026.02 | 18.8 | |
| Gemini-2.5-ProModel Type=Proprietary2026.02 | 21.9 | |
| InternVL2.5-38BModel Type=Open-Source, Parameters=38B2026.02 | 22.4 | |
| InternVL2.5-8BModel Type=Open-Source, Parameters=8B2026.02 | 29.7 | |
| Claude-4.5-HaikuModel Type=Proprietary2026.02 | 30.1 | |
| DeepSeek-VL2Model Type=Open-Source2026.02 | 38 |