List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC

97.5AccuracyGPT-5
50.90863.00475.187.196Aug 15, 2025Sep 19, 2025Oct 25, 2025Nov 30, 2025Jan 4, 2026Feb 9, 2026Mar 17, 2026
Updated 7mo ago

Evaluation results

MethodLinks
GPT-5Evaluation Protocol=Closed-Source2026.0197.5
Qwen3-235B-A22BModel Category=Large Language Models, Model Scale=235B-A22B, Reasoning Strategy=Thinking2025.1293.98
Qwen3-32BModel Category=Large Language Models, Model Scale=32B, Reasoning Strategy=Thinking2025.1293.98
FIGR2025.1293.98
GLM-4.5VModel Category=Large Vision-Language Models, Model Scale=108B2025.1287.95
Text-only RLMode=Text-only, Reasoning Strategy=RL2025.1287.95
Qwen3-VL-32B-InstructModel Category=Large Vision-Language Models, Model Scale=32B, Reasoning Strategy=Instruct2025.1284.34
TTRL-GTModel=Qwen2.5-7B-Base2026.0382.61
GPT-4.1Evaluation Protocol=Closed-Source2026.0182.5
ATLAS (cluster)Evaluation Protocol=In-Distribution2026.0182.5
Qwen3-VL-8B-InstructModel Category=Large Vision-Language Models, Model Scale=8B, Reasoning Strategy=Instruct2025.1281.93
PMPO-7B (Ours) [R1-Distill]Backbone=DeepSeek-R1-Distill-Qwen-7B, Optimization Algorithm=PMPO2026.0179.5
GMPO-7B [R1-Distill]Backbone=DeepSeek-R1-Distill-Qwen-7B, Optimization Algorithm=GMPO2026.0178.3
DCPOBackbone=Qwen2.5-math-7B, Optimization Method=DCPO, Sampling Strategy=x322026.0176.3
Dr.GRPO-7B [R1-Distill]Backbone=DeepSeek-R1-Distill-Qwen-7B, Optimization Algorithm=Dr.GRPO2026.0174.7
BAPORollouts=733k, Type=off, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0272.74
R-Few (5%)Backbone=Qwen3-8B-Base, Methodology=R-Few, Supervision Ratio=5%2025.1272.3
DistriTTRL-GMMModel=Qwen3-8B2026.0370.31
DAPORollouts=1921k, Type=on, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0270.08
SPICEBackbone=Qwen3-8B-Base, Methodology=SPICE, Supervision Ratio=None2025.1270
DCPOBackbone=Qwen2.5-7B, Optimization Method=DCPO, Sampling Strategy=x322026.0169.9
R-Few (1%)Backbone=Qwen3-8B-Base, Methodology=R-Few, Supervision Ratio=1%2025.1269.3
TTRL-WSCModel=Qwen3-8B2026.0368.8
PMPO-7B (Ours)Backbone=Qwen2.5-Math-7B, Optimization Algorithm=PMPO2026.0168.7
TTRL-GTModel=Qwen3-8B2026.0367.87
ATLAS (RL)Evaluation Protocol=Out-of-Distribution2026.0167.5
CAPO (RLOO)Backbone=Qwen2.5-7B-Math2025.1267.5
GRPORollouts=677k, Type=on, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0267.47
DistriTTRL-WSCModel=Qwen3-8B2026.0366.57
GSPO + LIEBackbone=Qwen3-4B-Base, Strategy=Length-Incentivized Exploration2026.0266.2
TTRLModel=Qwen3-8B2026.0365.95
GRPO w/Clip-higher + LIEBackbone=Qwen3-4B-Base, Variant=Clip-higher, Strategy=Length-Incentivized Exploration2026.0265.5
MoPPSRollouts=737k, Type=on, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0265.29
DEEP-GRPO-7BParameter size=7B2026.0265.1
GRPO (v = 5)Rollouts=677k, Type=off, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0265.09
Remix-GRPORollouts=N/A, Type=off, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0265.06
GPG-7BBackbone=Qwen2.5-Math-7B, Optimization Algorithm=GPG2026.0165
GPG-7BParameter size=7B2026.0265
CAPO (GRPO)Backbone=Qwen2.5-7B-Math2025.1265
General-ReasonerBackbone=Qwen3-8B-Base, Methodology=General-Reasoner, Supervision Ratio=232k WebInstruct2025.1264.8
RePORollouts=677k, Type=off, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0264.76
DeepSeek R1 Distill Qwen 1.5BRollouts=N/A, Type=N/A2026.0262.9
R-ZeroBackbone=Qwen3-8B-Base, Methodology=R-Zero, Supervision Ratio=None2025.1262.8
GSPOBackbone=Qwen3-4B-Base2026.0262.7
PRIME-Zero-7BBackbone=Qwen2.5-Math-7B, Optimization Algorithm=PRIME-Zero2026.0162.7
Eurus-7BBackbone=Mistral-7B, Optimization Algorithm=Eurus2026.0162.7
Dr.GRPO-7BBackbone=Qwen2.5-Math-7B, Optimization Algorithm=Dr.GRPO2026.0162.7
PRIME-Zero-7BParameter size=7B2026.0262.7
Eurus-7BParameter size=7B2026.0262.7
Oat-Zero-7B (Dr. GRPO)Parameter size=7B2026.0262.7
Absolute ZeroBackbone=Qwen3-8B-Base, Methodology=Absolute Zero, Supervision Ratio=None2025.1262.5
Gemini2.5-ProEvaluation Protocol=Closed-Source2026.0162.5
RouterDCEvaluation Protocol=In-Distribution2026.0162.5
CAPO (GRPO)Backbone=Qwen2.5-1.5B-Math2025.1262.5
CHORD-ϕmechanism=dual-control2025.0862.5
GRPO w/Clip-higherBackbone=Qwen3-4B-Base, Variant=Clip-higher2026.0261.8
Base ModelBackbone=Qwen3-8B-Base, Methodology=Base, Supervision Ratio=None2025.1261.5
Qwen3-32BModel Category=Large Language Models, Model Scale=32B, Reasoning Strategy=Non-Thinking2025.1261.45
GMPO-7BBackbone=Qwen2.5-Math-7B, Optimization Algorithm=GMPO2026.0161.4
DistriTTRL-GMMModel=Qwen2.5-Math-7B2026.0361.37
CHORD-µtransition=smooth2025.0860.8
GRPO + LIEBackbone=Qwen3-4B-Base, Strategy=Length-Incentivized Exploration2026.0260.5
SimpleRL-Zero-7BBackbone=Qwen2.5-Math-7B, Optimization Algorithm=SimpleRL-Zero2026.0160.2
SimpleRL-Zero-7BParameter size=7B2026.0260.2
General-ReasonerBackbone=Qwen3-4B-Base, Methodology=General-Reasoner, Supervision Ratio=232k WebInstruct2025.1260
TTRL-GTModel=Qwen2.5-Math-7B2026.0358.43
SFT-best + RLSFT intensity=thorough, RL=true2025.0858.4
TTRL-WSCModel=Qwen2.5-Math-7B2026.0358.28
DistriTTRL-WSCModel=Qwen2.5-Math-7B2026.0357.68
SPICEBackbone=Qwen3-4B-Base, Methodology=SPICE, Supervision Ratio=None2025.1257.5
CAPO (PPO)Backbone=Qwen2.5-7B-Math2025.1257.5
CAPO (PPO)Backbone=Qwen2.5-1.5B-Math2025.1257.5
CAPO (RLOO)Backbone=Qwen2.5-1.5B-Math2025.1257.5
DistriTTRL-WSCModel=Qwen2.5-7B-Base2026.0357.23
∇-ReasonerBackbone=Qwen-2.5-7B-Instruct, Nmax=82026.0356.8
DistriTTRL-GMMModel=Qwen2.5-7B-Base2026.0356.48
TTRLModel=Qwen2.5-Math-7B2026.0356.18
BoNBackbone=Qwen-2.5-7B-Instruct, N=82026.0355.9
TPOBackbone=Qwen-2.5-7B-Instruct2026.0355.9
SFT-bestSFT intensity=thorough2025.0855.9
DAREBackbone=Qwen3-1.7B, Method Category=Test-time scaling methods2026.0155.7
SCBackbone=Qwen-2.5-7B-Instruct, N=82026.0355.5
GRPOBackbone=Qwen3-4B-Base2026.0255.2
RLOOBackbone=Qwen2.5-7B-Math2025.1255
CAPO (Reinforce++)Backbone=Qwen2.5-7B-Math2025.1255
RAPBackbone=Qwen-2.5-7B-Instruct2026.0354.6
OpenReasoner-Zero-7B@8kBackbone=Qwen2.5-Math-7B, Optimization Algorithm=OpenReasoner-Zero, Sampling Budget=8k2026.0154.2
OpenReasoner-Zero-7B @ 8kParameter size=7B, Inference budget=8k2026.0254.2
TTRL-WSCModel=Qwen2.5-7B-Base2026.0354.07
SASR2025.0854
GRPOBackbone=Qwen3-1.7B, Method Category=RL methods on training dataset2026.0153.2
Dr.GRPO-1.5BBackbone=Qwen2.5-Math-1.5B, Optimization Algorithm=Dr.GRPO2026.0153
GMPO-1.5BBackbone=Qwen2.5-Math-1.5B, Optimization Algorithm=GMPO2026.0153
RLPRBackbone=Qwen3-1.7B, Method Category=Test-time scaling methods2026.0153
Oat-Zero-1.5B (Dr. GRPO)Parameter size=1.5B2026.0253
TTRLBackbone=Qwen3-1.7B, Method Category=Test-time scaling methods2026.0152.9
REINFORCE++Backbone=Qwen3-1.7B, Method Category=RL methods on training dataset2026.0152.8
GRPOBackbone=Qwen-2.5-7B2026.0352.8
LUFFY2025.0852.8
R-Few (1%)Backbone=Qwen3-4B-Base, Methodology=R-Few, Supervision Ratio=1%2025.1252.7