Visual Question Answering on BLINK (val)
73.7AccuracyGemini 2.5 Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini 2.5 ProAccess=API call only2026.03 | 73.7 | |
| Qwen3-VL-32B-Instruct + SpatialBoostEncoder Enhancement=SpatialBoost2026.03 | 70.8 | |
| Gemini 2.5 FlashAccess=API call only2026.03 | 70.2 | |
| InternVL3-38B + SpatialBoostEncoder Enhancement=SpatialBoost2026.03 | 69.2 | |
| Qwen3-VL-8BAccess=Open weights only2026.03 | 69.1 | |
| GPT-5 miniAccess=API call only2026.03 | 68.7 | |
| Qwen3-VL-32B-Instruct2026.03 | 67.3 | |
| GPT-5Access=API call only2026.03 | 66.5 | |
| Qwen3-VL-4BAccess=Open weights only2026.03 | 65.8 | |
| GLM-4.1V-9BAccess=Open weights only2026.03 | 65.1 | |
| Claude Sonnet 4.5Access=API call only2026.03 | 64.8 | |
| InternVL3-38B2026.03 | 64 | |
| InternVL3.5-8BAccess=Open weights only2026.03 | 59.5 | |
| InternVL3.5-4BAccess=Open weights only2026.03 | 58.1 | |
| Molmo2-4BAccess=Fully open2026.03 | 57.5 | |
| PLM-8BAccess=Fully open2026.03 | 56 | |
| PLM-3BAccess=Fully open2026.03 | 55.4 | |
| Keye-VL-1.5-8BAccess=Open weights only2026.03 | 54.9 | |
| Molmo2-8BAccess=Fully open2026.03 | 51.3 | |
| MolmoPoint-8BAccess=MolmoPoint2026.03 | 50.8 | |
| MolmoPoint-8B-O-7BAccess=Fully open2026.03 | 50.5 | |
| Eagle2.5-8BAccess=Open weights only2026.03 | 45.8 | |
| MRoPE-IPositional Encoding Method=MRoPE-I, DIPE Enhancement=+DIPE2026.03 | 42.45 | |
| MiniCPM-V-4.5-8BAccess=Open weights only2026.03 | 42 | |
| MRoPEPositional Encoding Method=MRoPE, DIPE Enhancement=+DIPE2026.03 | 41.93 | |
| Vanilla RoPEPositional Encoding Method=Vanilla RoPE, DIPE Enhancement=Base2026.03 | 40.93 | |
| MRoPE-IPositional Encoding Method=MRoPE-I, DIPE Enhancement=Base2026.03 | 40.29 | |
| MRoPEPositional Encoding Method=MRoPE, DIPE Enhancement=Base2026.03 | 39.87 | |
| Vanilla RoPEPositional Encoding Method=Vanilla RoPE, DIPE Enhancement=+DIPE2026.03 | 39.19 |