List your company
ResearchFormerly WizWand

Mathematical Reasoning on OlympiadBench (accuracy (%))

68.8AccuracySTARE-C2
36.788845.099453.4161.7206Jun 15, 2025Aug 15, 2025Oct 15, 2025Dec 15, 2025Feb 14, 2026Apr 16, 2026Jun 17, 2026
Updated 4mo ago

Evaluation results

MethodLinks
STARE-C2Scenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0668.8
JustRLScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0668
STARE-O1Scenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0667.8
GPT-o1-miniTool=✗2026.0265.3
GRPO-dsScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0664.1
STAPOScenario=Long CoT, Backbone Model=Qwen3-8B-Base, Model Scale=8B2026.0663
80/20 RuleScenario=Long CoT, Backbone Model=Qwen3-8B-Base, Model Scale=8B2026.0662.5
STARE-C2Scenario=Long CoT, Backbone Model=Qwen3-8B-Base, Model Scale=8B2026.0662.3
STARE-O1Scenario=Long CoT, Backbone Model=Qwen3-8B-Base, Model Scale=8B2026.0661.8
GRPOScenario=Long CoT, Backbone Model=Qwen3-8B-Base, Model Scale=8B2026.0661.5
Lp-RegScenario=Short CoT, Backbone Model=Qwen2.5-32B-Base, Model Scale=32B2026.0661.2
Fast-MathData Volume=7.9K2026.0361
STARE-C2Scenario=Short CoT, Backbone Model=Qwen2.5-32B-Base, Model Scale=32B2026.0660.9
EAPOScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0660.8
GRPOScenario=Short CoT, Backbone Model=Qwen2.5-32B-Base, Model Scale=32B2026.0660.3
STARE-O1Scenario=Short CoT, Backbone Model=Qwen2.5-32B-Base, Model Scale=32B2026.0660.1
Bidirectional Curriculum Generation (Round 4)Data Volume=5,8732026.0360.08
KL-CovScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0659.9
Bidirectional Curriculum Generation (Round 3)Data Volume=4,5942026.0359.35
EntroAdvScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0659.1
GRPO-dsScenario=Long CoT, Backbone Model=Qwen3-8B-Base, Model Scale=8B2026.0659
Raiden-DeepSeek-R1Data Volume=62K2026.0358.75
GRPOScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0658.7
DAPOScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0658.4
LaPha / sg@128Base model=Qwen2.5-Math-7B, Tool=✓2026.0258
MegaScienceData Volume=1.25M2026.0357.6
80/20 RuleScenario=Short CoT, Backbone Model=Qwen2.5-32B-Base, Model Scale=32B2026.0657.6
EntroRegScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0656.2
W-REINFScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0655.9
GSPOScenario=Short CoT, Backbone Model=Qwen2.5-32B-Base, Model Scale=32B2026.0655.6
DAPOScenario=Short CoT, Backbone Model=Qwen2.5-32B-Base, Model Scale=32B2026.0655.6
TTSRBackbone Model=Qwen3-8B-Base2026.0255.2
Base-1.5BScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0655.2
SimpleTIRScenario=Tool-Use Agent, Backbone Model=Qwen2.5-7B-Base, Model Scale=7B2026.0654.8
QWQ-LongCoTData Volume=130K2026.0354.15
GRPO-dsScenario=Short CoT, Backbone Model=Qwen2.5-32B-Base, Model Scale=32B2026.0654
STARE-C2Scenario=Tool-Use Agent, Backbone Model=Qwen2.5-7B-Base, Model Scale=7B2026.0653.6
STARE-O1Scenario=Tool-Use Agent, Backbone Model=Qwen2.5-7B-Base, Model Scale=7B2026.0652.3
Bidirectional Curriculum Generation (Round 2)Data Volume=1,6332026.0351.78
STARE-C2Scenario=Short CoT, Backbone Model=Qwen2.5-14B-Instruct, Model Scale=14B2026.0651.4
DeepSeek-R1-7BModel Scale=7B, Decoding Strategy=Greedy2026.0351
STARE-O1Scenario=Short CoT, Backbone Model=Qwen2.5-14B-Instruct, Model Scale=14B2026.0650.9
TTRLBackbone Model=Qwen3-8B-Base2026.0250.8
ToRLBase model=Qwen2.5-Math-7B, Tool=✓2026.0249.9
ToRLScenario=Tool-Use Agent, Backbone Model=Qwen2.5-7B-Base, Model Scale=7B2026.0649.9
TreePO / maj@16Base model=Qwen2.5-7B, Tool=✗2026.0249.2
Bidirectional Curriculum Generation (Round 1)Data Volume=8502026.0348.81
GRPO-dsScenario=Tool-Use Agent, Backbone Model=Qwen2.5-7B-Base, Model Scale=7B2026.0648.8
R-ZeroBackbone Model=Qwen3-8B-Base2026.0248.6
MATHFusionData Volume=74732026.0348.22
DGPOScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0648
FIM-TIESModel Scale=7B, Decoding Strategy=Greedy2026.0347.9
Qwen3-8B-baseData Volume=-2026.0347.2
LIMO2Data Volume=8002026.0346.74
HACPOModel Backbone=Qwen3-8B-Base2026.0346.7
ACM-TAModel Scale=7B, Decoding Strategy=Greedy2026.0346.7
ACM-TIESModel Scale=7B, Decoding Strategy=Greedy2026.0346.4
HACPOModel Backbone=Qwen3-4B-Base2026.0346.3
LaPha / sg@128Base model=Qwen2.5-7B, Tool=✓2026.0246
TTRLBackbone Model=Qwen3-4B-Base2026.0246
CISPOScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0645.8
CE-GPPO (Row 2)Scenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0645.6
80/20 RuleScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0645.5
TTSRBackbone Model=Qwen3-4B-Base2026.0245.3
STARE-C2Scenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0645.3
EAPOScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0645.1
SFTBase model=Qwen2.5-Math-7B, Tool=✓2026.0245
Task ArithmeticModel Scale=7B, Decoding Strategy=Greedy2026.0345
ASPOScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0644.9
STARE-O1Scenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0644.7
OpenO1-SFTData Volume=77K2026.0344.51
EntroAdvScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0644.5
FIM-TAModel Scale=7B, Decoding Strategy=Greedy2026.0344.3
SPECSBeam Width (n)=16, Backbone=Qwen2.5-Math-Instruct2025.0644.2
KL-CovScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0644.2
ToRLBase model=Qwen2.5-Math-1.5B, Tool=✓2026.0244
R-ZeroBackbone Model=Qwen3-4B-Base2026.0244
Sens-MergingModel Scale=7B, Decoding Strategy=Greedy2026.0343.9
DAPOScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0643.9
CE-GPPOScenario=Long CoT, Backbone Model=DeepSeek-R1-Distill-Qwen-1.5B, Model Scale=1.5B2026.0643.9
DAPOBase model=Qwen2.5-Math-7B, Tool=✗2026.0243.5
GRPO-dsScenario=Short CoT, Backbone Model=Qwen2.5-14B-Instruct, Model Scale=14B2026.0643.4
GPT-4oTool=✗2026.0243.3
STEERScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0643.3
TIES-MergingModel Scale=7B, Decoding Strategy=Greedy2026.0343
GRPOScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0642.8
EntroRegScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0642.8
DAPOBase model=Qwen2.5-Math-1.5B, Tool=✗2026.0242.7
GRPO-dsScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0642.6
PrimeBase model=Qwen2.5-Math-7B, Tool=✗2026.0242.1
SPECSBeam Width (n)=4, Backbone=Qwen2.5-Math-Instruct2025.0641.7
SFTBase model=Qwen2.5-Math-1.5B, Tool=✓2026.0241.3
Base ModelBackbone Model=Qwen3-8B-Base2026.0241.2
GSIBeam Width (n)=16, Backbone=Qwen2.5-Math-Instruct2025.0640.8
Base ModelBackbone Model=Qwen3-4B-Base2026.0240.2
LaPha / sg@128Base model=Qwen2.5-Math-1.5B, Tool=✓2026.0240
GSIBeam Width (n)=4, Backbone=Qwen2.5-Math-Instruct2025.0639.7
W-REINFScenario=Short CoT, Backbone Model=Qwen2.5-Math-7B-Base, Model Scale=7B2026.0638.9
DPPOModel=Qwen3-8B, Speedup=1.90×, rq=0.9, ro=0.92026.0338.06
DPPOModel=Qwen3-4B, Speedup=2.37×, rq=0.9, ro=0.92026.0338.02