Counting on Pixmo (val)
74AccuracyQwen2.5-VL
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VLSize=72B2026.02 | 74 | |
| Qwen2.5-VLSize=72B2026.06 | 74 | |
| Visual Para-Thinker++Size=7B2026.06 | 68.2 | |
| Visual Para-ThinkerSize=7B2026.02 | 65.7 | |
| Visual Para-ThinkerSize=7B2026.06 | 65.7 | |
| SequentialSize=7B2026.02 | 65.5 | |
| Sequential (SFT)Size=7B2026.06 | 65.5 | |
| GRPOSize=7B2026.06 | 64.4 | |
| Majority voting@4Size=7B2026.02 | 63.2 | |
| Majority voting@4Size=7B2026.06 | 63.2 | |
| Qwen2.5-VLSize=7B2026.02 | 63 | |
| Qwen2.5-VLSize=7B2026.06 | 63 | |
| Visual Para-Thinker++Size=3B2026.06 | 60.7 | |
| Gemini2.5-ProSize=-2026.06 | 59.8 | |
| Visual Para-ThinkerSize=3B2026.02 | 55 | |
| Visual Para-ThinkerSize=3B2026.06 | 55 | |
| GPT5-miniSize=-2026.06 | 54.7 | |
| GPT-4oSize=-2026.06 | 54.4 | |
| Claude-4-SonnetSize=-2026.06 | 53.5 | |
| GRPOSize=3B2026.06 | 53 | |
| Majority voting@4Size=3B2026.02 | 52.1 | |
| Majority voting@4Size=3B2026.06 | 52.1 | |
| SequentialSize=3B2026.02 | 51.5 | |
| Sequential (SFT)Size=3B2026.06 | 51.5 | |
| Qwen2.5-VLSize=3B2026.02 | 50.1 | |
| Qwen2.5-VLSize=3B2026.06 | 50.1 |