List your company
ResearchFormerly WizWand

Math Reasoning on OlympiadBench (Accuracy)

90.7AccuracyAPRM
-0.19623.4024770.598May 16, 2025Jun 29, 2025Aug 12, 2025Sep 25, 2025Nov 8, 2025Dec 22, 2025Feb 4, 2026
Updated 5mo ago

Evaluation results

MethodLinks
APRMBackbone=Gemma-3-27B2025.1190.7
APRMBackbone=GPT-OSS-20B2025.1190.4
APRMBackbone=GPT-OSS-120B2025.1189.4
ToTBackbone=GPT-OSS-20B2025.1185.6
ToTBackbone=Gemma-3-27B2025.1185.2
ReST-MCTSBackbone=GPT-OSS-20B2025.1184.7
ToTBackbone=GPT-OSS-120B2025.1184.3
ReST-MCTSBackbone=GPT-OSS-120B2025.1184.3
ReST-MCTSBackbone=Gemma-3-27B2025.1184.3
CoT-SCBackbone=GPT-OSS-120B2025.1183.9
LLM-JBackbone=GPT-OSS-20B2025.1183.9
LLM-JBackbone=GPT-OSS-120B2025.1183.1
LLM-JBackbone=Gemma-3-27B2025.1183.1
Hard Neg.Backbone=Gemma-3-27B2025.1183
Hard Neg.Backbone=GPT-OSS-120B2025.1179.2
ARMBackbone=GPT-OSS-120B2025.1177.6
CoT-SCBackbone=GPT-OSS-20B2025.1161
ARMBackbone=Gemma-3-27B2025.1161
TaH+Model Size=4B*2025.1152.6
OuroModel Size=4B*2025.1151.4
TaHModel Size=4B*2025.1150.5
StandardModel Size=4B*2025.1150.4
SoftThinkModel Size=4B*2025.1150.2
CoT-SCBackbone=Gemma-3-27B2025.1148.7
APRMBackbone=Gemma-3-12B2025.1144
CGDTeacher Model=S1.1-32B2025.0541.3
CGDTeacher Model=Claude Sonnet 3.72025.0541.2
ToTBackbone=Gemma-3-12B2025.1140.7
CFTTeacher Model=GPT-4o2025.0540.7
ReST-MCTSBackbone=Gemma-3-12B2025.1140.3
MoCANBackbone=Qwen2.5-7B-Instruct2025.1239.65
NSPOBackbone=Qwen2.5-7B-Instruct2025.1239
DPO-MixBackbone=Qwen2.5-7B-Instruct2025.1238.45
Qwen2.5-7B-Instruct2025.1237.8
LLM-JBackbone=Gemma-3-12B2025.1137.7
TaH+Model Size=1.7B2025.1137.6
TaHModel Size=1.7B2025.1137.2
CFTTeacher Model=S1.1-32B2025.0536.6
BFPOBackbone=Qwen2.5-7B-Instruct2025.1236.35
DPO-HBackbone=Qwen2.5-7B-Instruct2025.1236
PeCANBackbone=Qwen2.5-7B-Instruct2025.1235.95
DPO-SBackbone=Qwen2.5-7B-Instruct2025.1235.63
CoT-SCBackbone=Gemma-3-12B2025.1135.6
StandardModel Size=1.7B2025.1133.8
OuroModel Size=1.7B2025.1131.3
AlwaysThinkModel Size=1.7B2025.1131
SoftThinkModel Size=1.7B2025.1130.7
SafeRLHFBackbone=Qwen2.5-7B-Instruct2025.1222.22
TaH+Model Size=0.6B2025.1120.6
Qwen2.5-Math-7B (Base)2025.0519.9
TaHModel Size=0.6B2025.1119.1
StandardModel Size=0.6B2025.1115.4
W-DOORBackbone=Qwen2.5-7B-Instruct2025.1215.15
OuroModel Size=0.6B2025.1114.2
SoftThinkModel Size=0.6B2025.1114
AlwaysThinkModel Size=0.6B2025.1112.6
LycheeDecodeBase Model=DeepSeek-R1-Distill-Qwen-7B, Cache Correction=True2026.0212.5
TidalDecodeBase Model=DeepSeek-R1-Distill-Llama-8B, Cache Correction=False2026.0210.9
LycheeDecodeBase Model=DeepSeek-R1-Distill-Llama-8B, Cache Correction=False2026.0210.9
LycheeDecodeBase Model=DeepSeek-R1-Distill-Llama-8B, Cache Correction=True2026.0210.9
LycheeDecodeBase Model=DeepSeek-R1-Distill-Qwen-7B, Cache Correction=False2026.0210.9
Full AttentionBase Model=DeepSeek-R1-Distill-Llama-8B, Cache Correction=False2026.0210.2
Full AttentionBase Model=DeepSeek-R1-Distill-Qwen-7B, Cache Correction=False2026.0210.2
TidalDecodeBase Model=DeepSeek-R1-Distill-Llama-8B, Cache Correction=True2026.029.4
TidalDecodeBase Model=DeepSeek-R1-Distill-Qwen-7B, Cache Correction=True2026.028.6
TidalDecodeBase Model=DeepSeek-R1-Distill-Qwen-7B, Cache Correction=False2026.027
Llama3-8B-Instruct2025.125.7
BFPOBackbone=Llama3-8B-Instruct2025.125.49
W-DOORBackbone=Llama3-8B-Instruct2025.125.25
MoCANBackbone=Llama3-8B-Instruct2025.125.1
NSPOBackbone=Llama3-8B-Instruct2025.125
PeCANBackbone=Llama3-8B-Instruct2025.124.5
DPO-HBackbone=Llama3-8B-Instruct2025.123.85
DPO-MixBackbone=Llama3-8B-Instruct2025.123.8
DPO-SBackbone=Llama3-8B-Instruct2025.123.45
SafeRLHFBackbone=Llama3-8B-Instruct2025.123.3