List your company
ResearchFormerly WizWand

Mathematical Reasoning on DeepMath

66.2AccuracyRLVR (with verifier)
13.00426.814540.62554.4355May 20, 2025Jul 10, 2025Aug 30, 2025Oct 20, 2025Dec 10, 2025Jan 30, 2026Mar 22, 2026
Updated 4mo ago

Evaluation results

MethodLinks
RLVR (with verifier)Model size=7B, Reasoning budget=2048 tokens2025.1166.2
RAROModel size=7B, Reasoning budget=2048 tokens2025.1157.5
RLVR (with verifier)Model size=3B, Reasoning budget=2048 tokens2025.1155.8
RLVR (with verifier)Model size=1.5B, Reasoning budget=2048 tokens2025.1150.9
RL-LogitModel size=7B, Reasoning budget=2048 tokens, Notes=best over 2 variants2025.1149.3
RAROModel size=3B, Reasoning budget=2048 tokens2025.1149.1
RationalizationModel size=7B, Reasoning budget=2048 tokens2025.1148.6
PLR-4Backbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=46.362026.0346.36
PLR-EMABackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=322026.0346.16
PLR-1Backbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=322026.0345.62
Top-KBackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=322026.0345.13
BaseModel size=7B, Reasoning budget=2048 tokens2025.1144.2
StaticBackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=322026.0343.27
RL-LogitModel size=3B, Reasoning budget=2048 tokens, Notes=best over 2 variants2025.1143.1
PLR-1Backbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=162026.0342.69
PLR-EMABackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=162026.0342.52
PLR-4Backbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=162026.0342.46
SFTModel size=7B, Reasoning budget=2048 tokens2025.1142.3
Top-KBackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=162026.0342.2
StaticBackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=162026.0341.85
PLR-EMABackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=82026.0341.52
RAROModel size=1.5B, Reasoning budget=2048 tokens2025.1141.3
PLR-1Backbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=82026.0339.93
BaseModel size=3B, Reasoning budget=2048 tokens2025.1139.4
PLR-4Backbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=82026.0339.17
SFTModel size=3B, Reasoning budget=2048 tokens2025.1139
Top-KBackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=82026.0338.74
StaticBackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=82026.0338.7
RL-LogitModel size=1.5B, Reasoning budget=2048 tokens, Notes=best over 2 variants2025.1137.7
Iterative DPOModel size=7B, Reasoning budget=2048 tokens, Notes=max over 3 rounds2025.1136.9
PLR-EMABackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=42026.0336.64
PLR-1Backbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=42026.0336.64
PLR-4Backbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=42026.0336.64
Top-KBackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=42026.0336.49
StaticBackbone=Qwen2.5-7B-Instruct, k (number of ICL examples)=42026.0336.03
SFTModel size=1.5B, Reasoning budget=2048 tokens2025.1135.7
RationalizationModel size=1.5B, Reasoning budget=2048 tokens2025.1134.5
Iterative DPOModel size=3B, Reasoning budget=2048 tokens, Notes=max over 3 rounds2025.1134.2
Iterative DPOModel size=1.5B, Reasoning budget=2048 tokens, Notes=max over 3 rounds2025.1133
RationalizationModel size=3B, Reasoning budget=2048 tokens2025.1132.3
BaseModel size=1.5B, Reasoning budget=2048 tokens2025.1129.6
TATRA2026.0227.78
PIASTvariant=(E)2026.0225.33
PIAST2026.0224.85
PRL2026.0221.58
GPSvariant=SR-0.12026.0221.58
PRL2025.0521.58
GPSvariant=J2026.0221.4
GA2026.0218.63
GA2025.0518.63
PromptWizard2025.0518.52
PromptAgent2025.0516.43
DE2026.0216.1
DE2025.0516.1
APE2026.0215.47
APE2025.0515.47
GRACE2026.0215.05
GRACE2025.0515.05