Multimodal Question Answering on Frontier-comparable pool 20-eval macro mean (test)
10.23AME Change (pp)InternVL3-2B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| InternVL3-2BLength overlap=86%2026.06 | 10.23 | 0.27 | 0.21 | 0.8 | |
| InternVL3.5-2BLength overlap=26%2026.06 | 9.68 | 0.27 | 1.05 | 3.9 | |
| Qwen3.5-2BLength overlap=48%2026.06 | 6.99 | 0.27 | 0.97 | 3.6 | |
| InternVL3.5-4BLength overlap=53%2026.06 | 5.69 | 0.41 | 1.12 | 2.7 | |
| Qwen3-VL-4B-InstructLength overlap=81%2026.06 | 4.53 | 0.41 | 1.3 | 3.2 | |
| Qwen3-VL-2B-InstructLength overlap=89%2026.06 | 3.93 | 0.27 | 0.63 | 2.4 | |
| Perceptron-Isaac-2BLength overlap=18%2026.06 | 2.39 | 0.27 | 0.91 | 2.2 |