Abstract Visual Reasoning on VCog-Bench
42.4CVR ScoreOurs (RL)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ours (RL)Training=RL2026.01 | 42.4 | 12.1 | |
| Ours (SFT)Training=SFT2026.01 | 40.8 | 11.3 | |
| Qwen2.5-VL-7B-InstructParameters=7B, Variant=Instruct2026.01 | 39.8 | 12.1 | |
| InternVL2.5-8BParameters=8B2026.01 | 38.2 | 12.3 | |
| LLaVA-OneVision-Qwen2-7BParameters=7B2026.01 | 36.7 | 13 | |
| Janus-Pro-7BParameters=7B2026.01 | 34.5 | 12.5 | |
| MM-EurekaQwen-7BParameters=7B2026.01 | 33.3 | 10.4 | |
| Thyme2026.01 | 32.3 | 10.8 | |
| R1-onevision-RLTraining=RL2026.01 | 23.3 | 23.4 |