List your company
ResearchFormerly WizWand

Multimodal Reasoning on ZeroBench

26.35AccuracyNPO
18.716420.698222.6824.6618Apr 3, 2026Apr 13, 2026Apr 23, 2026May 3, 2026May 13, 2026May 23, 2026Jun 2, 2026
Updated 5mo ago

Evaluation results

MethodLinks
NPOstage=early-stage only2026.0426.35
GRPO + ReMindBase Model=Qwen3-VL-8B-Instruct2026.0626.05
DAPO + ReMindBase Model=Qwen3-VL-8B-Instruct2026.0626.05
SDPOBase Model=Qwen3-VL-8B-Instruct, Max Context Length=81922026.0425.15
RLSDBase Model=Qwen3-VL-8B-Instruct, Max Context Length=81922026.0424.85
NPOstage=early + late-stage2026.0424.85
AutoNPO2026.0424.7
RePOBase Model=Qwen3-VL-8B-Instruct2026.0623.5
GRPOBase Model=Qwen3-VL-8B-Instruct, Max Context Length=81922026.0422.6
GRPOtype=pure on-policy2026.0422.6
GRPOBase Model=Qwen3-VL-8B-Instruct2026.0622.6
GRPO + OPSDBase Model=Qwen3-VL-8B-Instruct, Max Context Length=81922026.0422.16
OPSDBase Model=Qwen3-VL-8B-Instruct, Max Context Length=81922026.0421.06
DAPOBase Model=Qwen3-VL-8B-Instruct2026.0620.66
LUFFYtype=external teacher2026.0420.51
Base LLMBase Model=Qwen3-VL-8B-Instruct, Max Context Length=81922026.0419.76
Qwen3-VL-8B-Instruct2026.0419.76
Base ModelBase Model=Qwen3-VL-8B-Instruct2026.0619.76
RLEPtype=far future2026.0419.61
RLEPBase Model=Qwen3-VL-8B-Instruct2026.0619.61
ExGRPOtype=historical replay2026.0419.01
ExGRPOBase Model=Qwen3-VL-8B-Instruct2026.0619.01