Faithful Reasoning on SPD-Faith Bench Multi-Difference 1.0 (test)
87.7CRGLM-4.5V
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GLM-4.5VAccess Type=Open-Source2026.02 | 87.7 | 58.3 | |
| Qwen3-VL-235B-A22BAccess Type=Open-Source2026.02 | 82.3 | 44.8 | |
| Gemini-2.5-ProAccess Type=Close-Source2026.02 | 78.2 | 40.4 | |
| GPT-4oAccess Type=Close-Source2026.02 | 74.8 | 39.3 | |
| Claude-4.5-HaikuAccess Type=Close-Source2026.02 | 65.4 | 38.2 | |
| Qwen3-VL-32BAccess Type=Open-Source2026.02 | 64.6 | 37.6 | |
| InternVL2.5-38BAccess Type=Open-Source2026.02 | 59.6 | 35.8 | |
| Qwen2.5-VL-72BAccess Type=Open-Source2026.02 | 56.8 | 31.7 | |
| DeepSeek-VL2Access Type=Open-Source2026.02 | 53.9 | 25.6 | |
| Qwen2.5-VL-7BAccess Type=Open-Source2026.02 | 38.2 | 29.5 | |
| MiniCPM-V-2.6Access Type=Open-Source2026.02 | 32.3 | 23.4 | |
| InternVL2.5-8BAccess Type=Open-Source2026.02 | 31 | 27.2 |