List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC 2023 (Accuracy and Average Tokens)

100AccuracyBCR-Qwen3-4B (Ours)
29.848.02566.2584.475Apr 2, 2026Apr 4, 2026Apr 7, 2026Apr 10, 2026Apr 12, 2026Apr 15, 2026Apr 18, 2026
Updated 6mo ago

Evaluation results

MethodLinks
BCR-Qwen3-4B (Ours)Inference mode=N=12026.041007,128
Qwen3-4B-Thinking-2507Inference mode=N=12026.0497.510,457
VanillaBackbone=Qwen3-8B2026.0497.58,648
GRPOBackbone=Qwen3-8B2026.0497.58,204
GRPO+SOPBackbone=Qwen3-8B2026.0497.55,431
VanillaBackbone=Qwen3-4B2026.0497.58,801
GRPOBackbone=Qwen3-4B2026.0497.57,712
GRPO-λBackbone=Qwen3-4B2026.0497.54,671
Step-GRPOBackbone=Qwen3-8B2026.04955,295
GRPO-8kBackbone=Qwen3-4B2026.04955,506
GRPO+LPBackbone=Qwen3-4B2026.04954,712
GRPO+SOPBackbone=Qwen3-4B2026.04955,793
Step-GRPOBackbone=Qwen3-4B2026.04955,456
GRPO+LPBackbone=Qwen3-8B2026.0492.53,786
DEER+SFTBackbone=Qwen3-8B2026.04906,149
GRPO-λBackbone=Qwen3-8B2026.04905,432
BCR-JustRL-1.5B (Ours)Inference mode=N=12026.0487.52,637
GRPO-8kBackbone=Qwen3-8B2026.0487.55,933
Qwen3-1.7BInference mode=N=12026.04858,157
e3-1.7BInference mode=N=1, Length Control=true2026.04857,154
JustRL-deepseek-1.5BInference mode=N=12026.04855,713
DEER+SFTBackbone=Qwen3-4B2026.04855,002
GRPOBackbone=Qwen3-1.7B2026.04859,425
GRPO+SOPBackbone=Qwen3-1.7B2026.0482.57,704
Step-GRPOBackbone=Qwen3-1.7B2026.0482.56,856
BroRL-1.5BInference mode=N=1, Length Control=true2026.04814,120
VanillaBackbone=Qwen3-1.7B2026.048010,713
GRPO-8kBackbone=Qwen3-1.7B2026.04807,021
GRPO-λBackbone=Qwen3-1.7B2026.04757,050
STILL-3-1.5BInference mode=N=12026.0472.55,706
GRPO+LPBackbone=Qwen3-1.7B2026.0472.56,605
DeepSeek-R1-Distill-Qwen-1.5BInference mode=N=12026.0467.59,702
DEER+SFTBackbone=Qwen3-1.7B2026.0467.55,712
ARM-3BInference mode=N=1, Adaptive Reasoning=true2026.0442.5703
Thinker-Q1.5BInference mode=N=1, Adaptive Reasoning=true2026.0432.5828