List your company
ResearchFormerly WizWand

Mathematical Reasoning on MATH-500 (Pass@1 Accuracy, Average Token Length)

97.4Pass@1 AccuracyBCR-Qwen3-4B (Ours)
30.11247.58165.0582.519Nov 13, 2025Dec 9, 2025Jan 5, 2026Feb 1, 2026Feb 28, 2026Mar 27, 2026Apr 23, 2026
Updated 5mo ago

Evaluation results

MethodLinks
BCR-Qwen3-4B (Ours)Inference mode=N=12026.0497.43,713
Qwen3-4B-Thinking-2507Inference mode=N=12026.04975,136
BroRL-1.5BInference mode=N=1, Length Control=true2026.0492.12,996
AutoThinkBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=7B2026.03922,378.2
LaserBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=7B2026.0391.81,648.3
TRiMSBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=7B2026.0391.4712.4
JustRL-deepseek-1.5BInference mode=N=12026.0491.43,099
VanillaBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=7B2026.03913,682.5
e3-1.7BInference mode=N=1, Length Control=true2026.04913,774
AdaptThinkBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=7B2026.0390.41,871.4
Qwen3-1.7BInference mode=N=12026.0490.24,621
JETBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=7B2026.0389.21,935.7
BCR-JustRL-1.5B (Ours)Inference mode=N=12026.0487.61,868
CoT Cold-Start + R-ZeroBackbone=Qwen3-8B-Base, # Train=64342025.1186.2—
AutoThinkBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=1.5B2026.0385.62,111.5
DeepSeek-R1-Distill-Qwen-1.5BInference mode=N=12026.0485.45,409
STILL-3-1.5BInference mode=N=12026.0484.64,280
LC-R1Base Model=DeepSeek-R1-Distill-Qwen, Model Scale=7B2026.0384.41,381.7
LaserBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=1.5B2026.0384.11,974.1
CoT Cold-Start + Solver FeedbackBackbone=Qwen3-8B-Base, # Train=68012025.1183.8—
TRiMSBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=1.5B2026.0383.4836.3
Seed SetBackbone=Qwen3-8B-Base, # Train=40002025.1183.4—
Self-InstructBackbone=Qwen3-8B-Base, # Train=59902025.1183.4—
Self-Instruct + R-ZeroBackbone=Qwen3-8B-Base, # Train=71422025.1183.4—
CoT Cold-Start + RLMTBackbone=Qwen3-8B-Base, # Train=63822025.1183.4—
Self-Instruct + Solver FeedbackBackbone=Qwen3-8B-Base, # Train=63332025.1182.8—
AdaptThinkBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=1.5B2026.0382.71,821.9
Self-Instruct + RLMTBackbone=Qwen3-8B-Base, # Train=71282025.1182.6—
VanillaBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=1.5B2026.0382.34,619.2
CoT Cold-Start + Solver FeedbackBackbone=Qwen3-4B-Base, # Train=67222025.1182—
CoT Cold-StartBackbone=Qwen3-8B-Base, # Train=67232025.1181.8—
Self-Instruct + Solver FeedbackBackbone=Qwen3-4B-Base, # Train=54892025.1181.2—
Self-Instruct + R-ZeroBackbone=Qwen3-4B-Base, # Train=64612025.1180.8—
CoT Cold-Start + RLMTBackbone=Qwen3-4B-Base, # Train=67032025.1180.8—
DDRLBackbone=Qwen2.5-Math-1.5B2026.0480.7—
Self-Instruct + CoT-Self-InstructBackbone=Qwen3-8B-Base, # Train=63882025.1180.2—
Seed SetBackbone=Qwen3-4B-Base, # Train=40002025.1179.6—
Self-Instruct + RLMTBackbone=Qwen3-4B-Base, # Train=68592025.1179.6—
CoT Cold-Start + R-ZeroBackbone=Qwen3-4B-Base, # Train=68652025.1178.4—
Self-Instruct + CoT-Self-InstructBackbone=Qwen3-4B-Base, # Train=70512025.1178.2—
CoT Cold-StartBackbone=Qwen3-4B-Base, # Train=66162025.1177.8—
Self-InstructBackbone=Qwen3-4B-Base, # Train=53752025.1177.4—
ETMRBackbone=Qwen2.5-Math-1.5B2026.0476.9—
TTRLBackbone=Qwen2.5-Math-1.5B2026.0473—
DDRLBackbone=Qwen2.5-Base-3B2026.0472.7—
TTRLBackbone=Qwen2.5-Base-3B2026.0472.2—
JETBase Model=DeepSeek-R1-Distill-Qwen, Model Scale=1.5B2026.03721,909.6
ETMRBackbone=Qwen2.5-Base-3B2026.0471.7—
LC-R1Base Model=DeepSeek-R1-Distill-Qwen, Model Scale=1.5B2026.0371.22,080.4
Base ModelBackbone=Qwen3-8B-Base, # Train=-2025.1167.8—
DDRLBackbone=Llama-3.1-8B-Instruct2026.0467.6—
Base ModelBackbone=Qwen3-4B-Base, # Train=-2025.1166.4—
TTRLBackbone=Llama-3.1-8B-Instruct2026.0463.7—
Thinker-Q1.5BInference mode=N=1, Adaptive Reasoning=true2026.0463.4760
ETMRBackbone=Llama-3.1-8B-Instruct2026.0459.5—
ARM-3BInference mode=N=1, Adaptive Reasoning=true2026.0458.41,222
No AdaptationBackbone=Qwen2.5-Base-3B2026.0453.2—
No AdaptationBackbone=Llama-3.1-8B-Instruct2026.0448.6—
No AdaptationBackbone=Qwen2.5-Math-1.5B2026.0432.7—