List your company
ResearchFormerly WizWand

Math Problem Solving on OlympiadBench

68.32AccuracyLESS
4.963221.411637.8654.3084Nov 11, 2025Dec 18, 2025Jan 24, 2026Mar 2, 2026Apr 8, 2026May 15, 2026Jun 21, 2026
Updated 4mo ago

Evaluation results

MethodLinks
LESSBase Model=OpenR1-Distill-7B2026.0668.32
DRIFTBase Model=OpenR1-Distill-7B2026.0668.29
QuratingBase Model=OpenR1-Distill-7B2026.0668.25
BM25Base Model=OpenR1-Distill-7B2026.0667.62
IF (Impl. w. GraSS)Base Model=OpenR1-Distill-7B2026.0667.62
OpenR1-Distill-7BBase Model=OpenR1-Distill-7B2026.0667.51
RDSBase Model=OpenR1-Distill-7B2026.0666.77
Random BaselineBase Model=OpenR1-Distill-7B2026.0666.58
DSIRBase Model=OpenR1-Distill-7B2026.0665.95
Self-DistillationBase Model=OpenR1-Distill-7B2026.0660.24
TaH+Param.=4B2025.1152
TaHParam.=4B2025.1150.5
SoftThinkParam.=4B2025.1150.1
StandardParam.=4B2025.1147.7
RoutingParam.=4B2025.1145
ACPORegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0642.83
ACPORegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0642.17
DAPORegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0641.44
TaH+Param.=1.7B2025.1141.2
CISPORegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0640.61
CISPORegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0639.1
TaHParam.=1.7B2025.1138.8
DAPORegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0637.2
Self-DistillationBase Model=Olmo3-7B-Instruct-SFT2026.0636.31
DRIFTBase Model=Olmo3-7B-Instruct-SFT2026.0635.61
BM25Base Model=Olmo3-7B-Instruct-SFT2026.0635.5
Random BaselineBase Model=Olmo3-7B-Instruct-SFT2026.0635.35
QuratingBase Model=Olmo3-7B-Instruct-SFT2026.0635.27
RDSBase Model=Olmo3-7B-Instruct-SFT2026.0634.9
Low-EntropyRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0634.69
High-EntropyRegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0634.54
IF (Impl. w. GraSS)Base Model=Olmo3-7B-Instruct-SFT2026.0634.31
Olmo3-7B-Instruct-SFTBase Model=Olmo3-7B-Instruct-SFT2026.0634.27
AR-LoptiRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0633.6
High-EntropyRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0633.54
SoftThinkParam.=1.7B2025.1133.4
StandardParam.=1.7B2025.1133
LESSBase Model=Olmo3-7B-Instruct-SFT2026.0632.34
RoutingParam.=1.7B2025.1130.6
AlwaysThinkParam.=1.7B2025.1130
Low-EntropyRegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0626.61
DSIRBase Model=Olmo3-7B-Instruct-SFT2026.0626.59
TaH+Param.=0.6B2025.1124
AR-LoptiRegime=OffP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0624
TaHParam.=0.6B2025.1123.9
SoftThinkParam.=0.6B2025.1119.4
StandardParam.=0.6B2025.1118.8
Base ModelRegime=N-OnP., Backbone=Qwen2.5-7B, Training Dataset=ORZ-57K, Sampling Strategy=avg@82026.0618.02
AlwaysThinkParam.=0.6B2025.1110.2
RoutingParam.=0.6B2025.117.4