Logic Reasoning on LogicVista
61.97LogicVista AccuracyQwen-8B-DeltaThinker
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen-8B-DeltaThinker2026.05 | 61.97 | — | |
| OPD w/ DELTAPROMPTS (10k)Training Data=DELTAPROMPTS (10k)2026.05 | 58.04 | — | |
| Qwen3-VL-8B-Thinking2026.05 | 56.64 | — | |
| OPD w/ Seed Data (10k)Training Data=Seed Data (10k)2026.05 | 56.24 | — | |
| Teachertype=RL2026.05 | 52.5 | 73.8 | |
| Uni-OPDframework=Multi-Teacher Distillation2026.05 | 52 | 73.8 | |
| OPDframework=Multi-Teacher Distillation2026.05 | 50 | 69.3 | |
| Studentmodel_base=Qwen3-VL-4B-Instruct2026.05 | 49.9 | 66.4 | |
| EASE-7BModel Scale=7B2026.05 | 49.6 | — | |
| VGPO-7BModel Scale=7B2026.05 | 49.4 | — | |
| VPPO-RL-7BModel Scale=7B2026.05 | 48.8 | — | |
| MM-Eureka-7BModel Scale=7B2026.05 | 47.9 | — | |
| VL-Rethinker-7BModel Scale=7B2026.05 | 47 | — | |
| NoisyRollout-7BModel Scale=7B2026.05 | 47 | — | |
| PAPO_D-7BModel Scale=7B2026.05 | 45.9 | — | |
| ThinkLite-VL-7BModel Scale=7B2026.05 | 44.3 | — |