Counting on Pixmo (test)
74AccuracyVisual Para-Thinker++
Evaluation Results
| Method | Links | |
|---|---|---|
| Visual Para-Thinker++Size=7B2026.06 | 74 | |
| Visual Para-ThinkerSize=7B2026.02 | 70.8 | |
| Visual Para-ThinkerSize=7B2026.06 | 70.8 | |
| Qwen2.5-VLSize=72B2026.02 | 70.4 | |
| Qwen2.5-VLSize=72B2026.06 | 70.4 | |
| Visual Para-Thinker++Size=3B2026.06 | 69.5 | |
| GRPOSize=7B2026.06 | 69.2 | |
| SequentialSize=7B2026.02 | 68 | |
| Sequential (SFT)Size=7B2026.06 | 68 | |
| Qwen2.5-VLSize=7B2026.02 | 67.7 | |
| Qwen2.5-VLSize=7B2026.06 | 67.7 | |
| Majority voting@4Size=7B2026.02 | 67.6 | |
| Majority voting@4Size=7B2026.06 | 67.6 | |
| Gemini2.5-ProSize=-2026.02 | 59.8 | |
| GPT5-miniSize=-2026.02 | 54.7 | |
| GPT-4oSize=-2026.02 | 54.4 | |
| Visual Para-ThinkerSize=3B2026.02 | 54.4 | |
| Visual Para-ThinkerSize=3B2026.06 | 54.4 | |
| Claude-4-SonnetSize=-2026.02 | 53.5 | |
| GRPOSize=3B2026.06 | 52.9 | |
| Majority voting@4Size=3B2026.02 | 52.8 | |
| Majority voting@4Size=3B2026.06 | 52.8 | |
| SequentialSize=3B2026.02 | 52.1 | |
| Sequential (SFT)Size=3B2026.06 | 52.1 | |
| Qwen2.5-VLSize=3B2026.02 | 51.6 | |
| Qwen2.5-VLSize=3B2026.06 | 51.6 |