List your company
ResearchFormerly WizWand

Mathematical Reasoning on Olympiad

70.9AccuracyDeepSeek-R1-Distill-Qwen-7B + ERC-DAPO
31.69241.87152.0562.229Jun 30, 2025Aug 13, 2025Sep 26, 2025Nov 10, 2025Dec 24, 2025Feb 6, 2026Mar 23, 2026
Updated 7mo ago

Evaluation results

MethodLinks
DeepSeek-R1-Distill-Qwen-7B + ERC-DAPOModel Backbone=Qwen-7B, Training Algorithm=ERC-DAPO2025.1270.9
DeepSeek-R1-Distill-Qwen-7B + DAPOModel Backbone=Qwen-7B, Training Algorithm=DAPO2025.1269.9
DeepSeek-R1-Distill-Qwen-7BModel Backbone=Qwen-7B, Training Algorithm=Baseline2025.1267
DeepSeek-R1-Distill-Qwen-7B + GRPOModel Backbone=Qwen-7B, Training Algorithm=GRPO2025.1265.6
DeepSeek-R1-Distill-Qwen-1.5B + ERC-DAPOModel Backbone=Qwen-1.5B, Training Algorithm=ERC-DAPO2025.1261
SOUPratioBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Truncation=Length-ratio-based2026.0159.05
DeepSeek-R1-Distill-Qwen-1.5B + DAPOModel Backbone=Qwen-1.5B, Training Algorithm=DAPO2025.1258.6
VanillaModel=Qwen3-8B, Budget=24002026.0358.22
VanillaModel=Qwen3-8B, Budget=32002026.0358.22
On-policyBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0158
SPIRALBackbone=DeepSeek-Distill-Qwen-7B, Training Data=Multi-game2025.0657.9
GSPO + LIEBackbone=Qwen3-4B-Base, Strategy=Length-Incentivized Exploration2026.0257.2
M2POBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0156.92
DeepSeek-Distill-Qwen-7BBackbone=DeepSeek-Distill-Qwen-7B2025.0656.9
DeepSeek-R1-Distill-Qwen-1.5B + GRPOModel Backbone=Qwen-1.5B, Training Algorithm=GRPO2025.1256.2
R-KVModel=Qwen3-8B, Budget=32002026.0355.56
LongFlowModel=Qwen3-8B, Budget=32002026.0355.56
VATPModel=Qwen3-8B, Budget=32002026.0355.41
GRPO w/Clip-higherBackbone=Qwen3-4B-Base, Variant=Clip-higher2026.0254.1
GRPO w/Clip-higher + LIEBackbone=Qwen3-4B-Base, Variant=Clip-higher, Strategy=Length-Incentivized Exploration2026.0254.1
H2OModel=Qwen3-8B, Budget=32002026.0353.78
R-KVModel=Qwen3-8B, Budget=24002026.0353.33
SFTBackbone=DeepSeek-Distill-Qwen-7B, Training Data=Multi-game2025.0652.4
SOUPentropyBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Truncation=Entropy-based2026.0152.2
DeepSeek-R1-Distill-Qwen-1.5BModel Backbone=Qwen-1.5B, Training Algorithm=Baseline2025.1251.8
GSPOBackbone=Qwen3-4B-Base2026.0251.7
LUFFYBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0151.61
LongFlowModel=Qwen3-8B, Budget=24002026.0351.11
DPO-R1 (HIGH)Backbone=Qwen3-4B-Instruct, Compute Cost (Sampling Budget N)=N = 16, Trainable Pairs=8,964, Processing Time=5.22026.0250.8
PACEBackbone=Qwen3-4B-Instruct, Compute Cost (Sampling Budget N)=2 < N < 3, Trainable Pairs=10,717, Processing Time=1.02026.0250.8
DPO-R1 (LOW)Backbone=Qwen3-4B-Instruct, Compute Cost (Sampling Budget N)=N = 2, Trainable Pairs=3,479, Processing Time=0.92026.0250.7
DPO-R1 (ZHANG ET AL., 2025)Backbone=Qwen3-8B-Instruct, Compute Cost (Sampling Budget N)=N = 8, Trainable Pairs=10,211, Processing Time=4.82026.0250.7
DPO-R1 (MIDDLE)Backbone=Qwen3-8B-Instruct, Compute Cost (Sampling Budget N)=N = 4, Trainable Pairs=6,073, Processing Time=2.22026.0250.5
DPO-R1 (HIGH)Backbone=Qwen3-8B-Instruct, Compute Cost (Sampling Budget N)=N = 16, Trainable Pairs=10,543, Processing Time=7.22026.0250.5
DPO-R1 (MIDDLE)Backbone=Qwen3-4B-Instruct, Compute Cost (Sampling Budget N)=N = 4, Trainable Pairs=5,127, Processing Time=1.22026.0250.4
LUFFYBackbone=Qwen2.5-Math-7B2026.0149.99
GRPO + LIEBackbone=Qwen3-4B-Base, Strategy=Length-Incentivized Exploration2026.0249.9
INSTRUCT (BASE)Backbone=Qwen3-8B-Instruct, Compute Cost (Sampling Budget N)=N/A, Trainable Pairs=N/A, Processing Time=N/A2026.0249.9
DPO-R1 (LOW)Backbone=Qwen3-8B-Instruct, Compute Cost (Sampling Budget N)=N = 2, Trainable Pairs=3,403, Processing Time=1.42026.0249.9
PACEBackbone=Qwen3-8B-Instruct, Compute Cost (Sampling Budget N)=2 < N < 3, Trainable Pairs=9,028, Processing Time=1.72026.0249.6
SPIRALBackbone=Qwen3-8B, Training Data=Multi-game2025.0649.6
VATPModel=Qwen3-8B, Budget=24002026.0349.33
H2OModel=Qwen3-8B, Budget=24002026.0349.04
DPO-R1 (ZHANG ET AL., 2025)Backbone=Qwen3-4B-Instruct, Compute Cost (Sampling Budget N)=N = 8, Trainable Pairs=8,403, Processing Time=2.82026.0248.9
SOUPratioBackbone=Qwen2.5-Math-7B, Truncation=Length-ratio-based2026.0148.32
EASDTM=32B, DM=7B2025.1248.15
Absolute ZeroBackbone=Qwen3-8B-Base, Methodology=Absolute Zero, Supervision Ratio=None2025.1247.8
General-ReasonerBackbone=Qwen3-4B-Base, Methodology=General-Reasoner, Supervision Ratio=232k WebInstruct2025.1247.7
INSTRUCT (BASE)Backbone=Qwen3-4B-Instruct, Compute Cost (Sampling Budget N)=N/A, Trainable Pairs=N/A, Processing Time=N/A2026.0247.4
GRPOBackbone=Qwen3-4B-Base2026.0247.1
Single ModelTM=32B2025.1246.81
R-KVModel=DeepSeek-R1-Distill-Llama-8B, Budget=32002026.0346.81
VATPModel=DeepSeek-R1-Distill-Llama-8B, Budget=32002026.0346.67
DPO (Full)Base Model=Qwen2.5-7B-Instruct2026.0246.5
RSDTM=32B, DM=7B, PRM=1.5B2025.1246.5
R-Few (5%)Backbone=Qwen3-8B-Base, Methodology=R-Few, Supervision Ratio=5%2025.1246.4
General-ReasonerBackbone=Qwen3-8B-Base, Methodology=General-Reasoner, Supervision Ratio=232k WebInstruct2025.1246.3
H2OModel=DeepSeek-R1-Distill-Llama-8B, Budget=32002026.0346.22
SOUPentropyBackbone=Qwen2.5-Math-7B, Truncation=Entropy-based2026.0146.14
On-policyBackbone=Qwen2.5-Math-7B2026.0145.97
R-KVModel=DeepSeek-R1-Distill-Llama-8B, Budget=24002026.0345.93
LongFlowModel=DeepSeek-R1-Distill-Llama-8B, Budget=32002026.0345.93
SFTBackbone=Qwen3-8B, Training Data=Multi-game2025.0645.9
RSDTM=72B, DM=7B, PRM=1.5B2025.1245.62
SAGEBase Model=Qwen2.5-7B-Instruct2026.0245.5
Majority Voting (N=16)DM=7B2025.1245.33
SDTM=32B, DM=7B2025.1245.33
VanillaBase Model=Qwen2.5-7B-Instruct2026.0245.3
EASDTM=72B, DM=7B2025.1244.74
Single ModelTM=72B2025.1244.59
VanillaModel=DeepSeek-R1-Distill-Llama-8B, Budget=24002026.0344.3
VanillaModel=DeepSeek-R1-Distill-Llama-8B, Budget=32002026.0344.3
SDTM=72B, DM=7B2025.1244.15
R-Few (1%)Backbone=Qwen3-8B-Base, Methodology=R-Few, Supervision Ratio=1%2025.1244
H2OModel=DeepSeek-R1-Distill-Llama-8B, Budget=24002026.0344
LongFlowModel=DeepSeek-R1-Distill-Llama-8B, Budget=24002026.0343.85
M2POBackbone=Qwen2.5-Math-7B2026.0143.76
VATPModel=DeepSeek-R1-Distill-Llama-8B, Budget=24002026.0343.56
R-ZeroBackbone=Qwen3-8B-Base, Methodology=R-Zero, Supervision Ratio=None2025.1243.4
DPO (Random)Base Model=Qwen2.5-7B-Instruct2026.0243
R-Few (5%)Backbone=Qwen3-4B-Base, Methodology=R-Few, Supervision Ratio=5%2025.1242.8
SPICEBackbone=Qwen3-4B-Base, Methodology=SPICE, Supervision Ratio=None2025.1242.7
SPICEBackbone=Qwen3-8B-Base, Methodology=SPICE, Supervision Ratio=None2025.1242.5
R-Few (1%)Backbone=Qwen3-4B-Base, Methodology=R-Few, Supervision Ratio=1%2025.1242.4
SPIRALBackbone=Qwen3-4B, Training Data=Multi-game2025.0641.8
Absolute ZeroBackbone=Qwen3-4B-Base, Methodology=Absolute Zero, Supervision Ratio=None2025.1241.5
R-ZeroBackbone=Qwen3-4B-Base, Methodology=R-Zero, Supervision Ratio=None2025.1240.6
Base ModelBackbone=Qwen3-8B-Base, Methodology=Base, Supervision Ratio=None2025.1240.4
Beam Search (N=16)DM=7B2025.1240.15
SPIRALBackbone=Qwen3-4B, Training Data=Kuhn Poker2025.0638.4
SFTBackbone=Qwen3-4B, Training Data=Multi-game2025.0637.6
SFTBackbone=Qwen3-4B, Training Data=Kuhn Poker2025.0636.7
FIM-TIESModel Scale=1.5B, Avg. Length=411, Random Seeds=42026.0336.3
Single ModelDM=7B2025.1235.7
Base ModelBackbone=Qwen3-4B-Base, Methodology=Base, Supervision Ratio=None2025.1234.8
ACM-TIESModel Scale=1.5B, Avg. Length=1,4892026.0333.8
SPIRALBackbone=Octothinker-8B, Training Data=Multi-game2025.0633.7
Qwen3-8B-BaseBackbone=Qwen3-8B2025.0633.5
Qwen3-4B-BaseBackbone=Qwen3-4B2025.0633.3
Qwen3-4B-Base2026.0233.2