Counting on CountBench (test)
91.5AccuracyDr. Seg
Evaluation Results
| Method | Links | |
|---|---|---|
| Dr. Seg2026.02 | 91.5 | |
| Claude-4-SonnetSize=-2026.06 | 90.2 | |
| Gemini2.5-ProSize=-2026.06 | 89.7 | |
| Visual Para-Thinker++Size=7B2026.06 | 88.3 | |
| VisionReasoner2026.02 | 87.6 | |
| Qwen2.5-VLSize=72B2026.06 | 86.7 | |
| GPT-4oSize=-2026.06 | 85.7 | |
| Visual Para-ThinkerSize=7B2026.06 | 85.4 | |
| GPT5-miniSize=-2026.06 | 84.4 | |
| GRPOSize=7B2026.06 | 84.2 | |
| Sequential (SFT)Size=7B2026.06 | 83.7 | |
| Majority voting@4Size=7B2026.06 | 83.2 | |
| Qwen2.5-VLSize=7B2026.06 | 83.1 | |
| Qwen2-VL2026.02 | 76.5 | |
| Qwen2.5-VL2026.02 | 76 | |
| Visual Para-Thinker++Size=3B2026.06 | 74.8 | |
| Visual Para-ThinkerSize=3B2026.06 | 72.6 | |
| GRPOSize=3B2026.06 | 71.2 | |
| Sequential (SFT)Size=3B2026.06 | 70.4 | |
| Majority voting@4Size=3B2026.06 | 69.8 | |
| Qwen2.5-VLSize=3B2026.06 | 69.2 |