Science Question Answering on ScienceQA (Calibration/Ranking Metrics)
0.88ECEInternVL2.5-4B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| InternVL2.5-4BConf.=Token Confidence, Source=Open-source MLLMs2026.04 | 0.88 | 92.81 | 99.95 | 92.57 | |
| MiniCPM-V-2.6Conf.=Token Confidence, Source=Open-source MLLMs2026.04 | 1.88 | 98.69 | 99.99 | 98.69 | |
| Phi-3.5-VisionConf.=Token Confidence, Source=Open-source MLLMs2026.04 | 2.69 | 92.43 | 99.46 | 91.75 | |
| GPT-4oConf.=Verbal Confidence, Source=Closed-source MLLMs2026.04 | 4.25 | 91.71 | 99.07 | 91.02 | |
| InternVL3.5-8BConf.=Token Confidence, Source=Open-source MLLMs2026.04 | 4.53 | 98.67 | 99.88 | 98.68 | |
| Qwen2.5-VL-7BConf.=Token Confidence, Source=Open-source MLLMs2026.04 | 4.76 | 99.25 | 99.92 | 99.2 | |
| InternVL2.5-4BConf.=Verbal Confidence, Source=Open-source MLLMs2026.04 | 7.64 | 79.41 | 99.87 | 79.87 | |
| GPT-4oConf.=Token Confidence, Source=Closed-source MLLMs2026.04 | 7.82 | 93.48 | 99.25 | 92.79 | |
| GPT-4o-miniConf.=Token Confidence, Source=Closed-source MLLMs2026.04 | 7.93 | 92.55 | 99.26 | 91.88 | |
| InternVL3.5-8BConf.=Verbal Confidence, Source=Open-source MLLMs2026.04 | 9.09 | 87.73 | 98.62 | 84.79 | |
| Qwen2.5-VL-7BConf.=Verbal Confidence, Source=Open-source MLLMs2026.04 | 12.91 | 69.17 | 95.6 | 55.37 | |
| MiniCPM-V-2.6Conf.=Verbal Confidence, Source=Open-source MLLMs2026.04 | 14.9 | 52.61 | 99.64 | 45.05 | |
| GPT-4o-miniConf.=Verbal Confidence, Source=Closed-source MLLMs2026.04 | 16.1 | 90.03 | 98.98 | 88.75 | |
| Phi-3.5-VisionConf.=Verbal Confidence, Source=Open-source MLLMs2026.04 | 36.86 | 66.91 | 96.76 | 50.14 |