List your company

Subjective Evaluation on WildBench

0.8604ScoreSTEP3-VL-10B
0.195320.3679850.540650.713315Jan 14, 2026Jan 28, 2026Feb 11, 2026Feb 25, 2026Mar 11, 2026Mar 25, 2026Apr 9, 2026

Evaluation results

MethodLinks
STEP3-VL-10BNumber of Parameters=10B2026.010.8604
Qwen3-VL ThinkingNumber of Parameters=8B2026.010.7236
MiMo-VL RL-2508Number of Parameters=7B2026.010.6309
InternVL 3.5Number of Parameters=8B2026.010.5645
+ GRPOavgBackbone=Qwen3-8B2026.040.5589
+ GRPOimpBackbone=Qwen3-8B2026.040.5573
SPARDBackbone=Qwen3-8B2026.040.5529
BaseBackbone=Qwen3-8B2026.040.5521
+ GRPOrmBackbone=Qwen3-8B2026.040.551
+ DPOBackbone=Qwen3-8B2026.040.5473
+ GRPOavgBackbone=Qwen2.5-7B-Instruct2026.040.4508
SPARDBackbone=Qwen2.5-7B-Instruct2026.040.4485
+ DPOBackbone=Qwen2.5-7B-Instruct2026.040.436
+ GRPOimpBackbone=Qwen2.5-7B-Instruct2026.040.4247
+ GRPOrmBackbone=Qwen2.5-7B-Instruct2026.040.4177
BaseBackbone=Qwen2.5-7B-Instruct2026.040.4172
GLM-4.6V FlashNumber of Parameters=9B2026.010.3404
+ SFTBackbone=Qwen2.5-7B-Instruct2026.040.2405
+ SFTBackbone=Qwen3-8B2026.040.2209