Subjective Evaluation on WildBench
0.8604ScoreSTEP3-VL-10B
Evaluation Results
| Method | Links | |
|---|---|---|
| STEP3-VL-10BNumber of Parameters=10B2026.01 | 0.8604 | |
| Qwen3-VL ThinkingNumber of Parameters=8B2026.01 | 0.7236 | |
| MiMo-VL RL-2508Number of Parameters=7B2026.01 | 0.6309 | |
| InternVL 3.5Number of Parameters=8B2026.01 | 0.5645 | |
| + GRPOavgBackbone=Qwen3-8B2026.04 | 0.5589 | |
| + GRPOimpBackbone=Qwen3-8B2026.04 | 0.5573 | |
| SPARDBackbone=Qwen3-8B2026.04 | 0.5529 | |
| BaseBackbone=Qwen3-8B2026.04 | 0.5521 | |
| + GRPOrmBackbone=Qwen3-8B2026.04 | 0.551 | |
| + DPOBackbone=Qwen3-8B2026.04 | 0.5473 | |
| + GRPOavgBackbone=Qwen2.5-7B-Instruct2026.04 | 0.4508 | |
| SPARDBackbone=Qwen2.5-7B-Instruct2026.04 | 0.4485 | |
| + DPOBackbone=Qwen2.5-7B-Instruct2026.04 | 0.436 | |
| + GRPOimpBackbone=Qwen2.5-7B-Instruct2026.04 | 0.4247 | |
| + GRPOrmBackbone=Qwen2.5-7B-Instruct2026.04 | 0.4177 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.04 | 0.4172 | |
| GLM-4.6V FlashNumber of Parameters=9B2026.01 | 0.3404 | |
| + SFTBackbone=Qwen2.5-7B-Instruct2026.04 | 0.2405 | |
| + SFTBackbone=Qwen3-8B2026.04 | 0.2209 |