List your company
ResearchFormerly WizWand

Mathematical Reasoning on Olympiad Math (Accuracy)

61.5AccuracySCOPE-RL
7.4221.4635.549.54Oct 9, 2025Nov 11, 2025Dec 14, 2025Jan 16, 2026Feb 18, 2026Mar 23, 2026Apr 26, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SCOPE-RLBase Model=Qwen3-4B2025.1061.5
DAPOBase Model=Qwen3-4B2025.1061.1
CISPOBase Model=Qwen3-4B2025.1060.3
Entropy-RegBase Model=Qwen3-4B, λ=0.012025.1060.1
GRPOBase Model=Qwen3-4B2025.1060
KL-covBase Model=Qwen3-4B2025.1060
Entropy-AdvBase Model=Qwen3-4B2025.1058.4
SFT→RLBase Model=Llama-3.1-8B, Chat Template=Corrected Baseline2026.0448.5
Qwen3-4BBase Model=Qwen3-4B2025.1047.2
SCOPE-RLBase Model=Qwen2.5-Math-7B2025.1043
Entropy-AdvBase Model=Qwen2.5-Math-7B2025.1042.1
GRPOBase Model=Qwen2.5-Math-7B2025.1041.8
KL-covBase Model=Qwen2.5-Math-7B2025.1041.7
CISPOBase Model=Qwen2.5-Math-7B2025.1041.6
Entropy-RegBase Model=Qwen2.5-Math-7B, λ=0.012025.1040.8
DAPOBase Model=Qwen2.5-Math-7B2025.1040.8
Entropy-RegBase Model=Qwen2.5-Math-7B, λ=0.032025.1040.4
Clip higherBase Model=Qwen2.5-Math-7B, ϵhigh=0.42025.1040.4
SCOPE-RLBase Model=Qwen2.5-7B2025.1040.3
DAPOBase Model=Qwen2.5-7B2025.1040.1
Entropy-RegBase Model=Qwen2.5-Math-7B, λ=0.052025.1039.2
Entropy-RegBase Model=Qwen2.5-7B, λ=0.012025.1039.1
KL-covBase Model=Qwen2.5-7B2025.1039.1
Clip higherBase Model=Qwen2.5-Math-7B, ϵhigh=0.52025.1038.9
GRPOBase Model=Qwen2.5-7B2025.1038.8
CISPOBase Model=Qwen2.5-7B2025.1038.2
Entropy-AdvBase Model=Qwen2.5-7B2025.1037.5
SFTBase Model=Llama-3.1-8B, Chat Template=Corrected Baseline2026.0436.2
Qwen2.5-7BBase Model=Qwen2.5-7B2025.1029.3
Qwen2.5-Math-7BBase Model=Qwen2.5-Math-7B2025.1026.7
HPTBase Model=Llama-3.1-8B, Chat Template=Paper-Original2026.0420.4
ReLIFTBase Model=Llama-3.1-8B2026.0412.8
Prefix-RFTBase Model=Llama-3.1-8B, Chat Template=Paper-Original2026.0411.9
LUFFYBase Model=Llama-3.1-8B2026.0410.8
SRFTBase Model=Llama-3.1-8B, Chat Template=Paper-Original2026.049.5