List your company

Graduate QA on GPQA (held-out)

31.5Accuracy (GPQA held-out)Baseline
16.26420.219524.17528.1305Jun 10, 2026

Evaluation results

MethodLinks
BaselineModel=Qwen3.5-2B2026.0631.5
Random StringModel=Qwen3.5-2B2026.0630.77
LoRA (DAPO)Model=Qwen3.5-2B2026.0630.04
Random ImageModel=Qwen3.5-2B2026.0627.47
Optimized Image (ART)Model=Qwen3.5-2B2026.0626.37
LoRA (DAPO)Model=Qwen3.5-0.8B2026.0624.18
Fixed Seed ImageModel=Qwen3.5-2B2026.0624.18
BaselineModel=Qwen3.5-0.8B2026.0623.44
Optimized Image (ART)Model=Qwen3.5-0.8B2026.0620.15
Random ImageModel=Qwen3.5-0.8B2026.0619.78
Random StringModel=Qwen3.5-0.8B2026.0619.41
Fixed Seed ImageModel=Qwen3.5-0.8B2026.0616.85