List your company
ResearchFormerly WizWand

Math Reasoning on GSM8K (Accuracy)

98.87AccuracyDMoA
52.652464.651276.6588.6488Feb 5, 2025Apr 23, 2025Jul 9, 2025Sep 25, 2025Dec 11, 2025Feb 26, 2026May 15, 2026
Updated 5mo ago

Evaluation results

MethodLinks
DMoABackbone=gpt-oss-120b, Evaluation Paradigm=few-shot2026.0598.87
DCCDModel Size=14B, Backbone=Qwen 2.5 14B, Algorithm=Draft Conditioned Constrained Decoding2026.0295.15
ARG-DesignerBackbone=gpt-oss-120b, Evaluation Paradigm=few-shot2026.0594.4
STEERBackbone=gpt-oss-120b2026.0593.5
G-DesignerBackbone=gpt-oss-120b, Evaluation Paradigm=few-shot2026.0593.35
SafeSieveBackbone=gpt-oss-120b2026.0593.2
MAGEBackbone=Llama-3.1-8B, Execution Scaffold=Sonnet, Number of seeds=52026.0592.5
MAGEBackbone=Llama-3.1-8B, Execution Scaffold=Opus, Number of seeds=52026.0592
SpecReasonBackbone=gpt-oss-120b2026.0591.4
DCCDModel Size=7B, Backbone=Qwen 2.5 7B, Algorithm=Draft Conditioned Constrained Decoding2026.0291.28
CPModel Size=14B, Backbone=Qwen 2.5 14B, Algorithm=Constrained Prompting2026.0291.13
CFModel Size=14B, Backbone=Qwen 2.5 14B, Algorithm=Constrained Few Shot2026.0290.52
GPTSwarmBackbone=gpt-oss-120b2026.0590.2
LLM-DebateBackbone=gpt-oss-120b2026.0589.95
AFlowBackbone=gpt-oss-120b2026.0589.5
AutoGenBackbone=gpt-oss-120b2026.0589.4
TeacherModel=Qwen2.5-7B-Instruct, Role=Teacher2025.1089.3
MoABackbone=gpt-oss-120b2026.0588.52
Random GraphBackbone=gpt-oss-120b2026.0588.25
Complete GraphBackbone=gpt-oss-120b2026.0588.05
SCBackbone=gpt-oss-120b2026.0587.66
CoTBackbone=gpt-oss-120b2026.0587.35
ChainBackbone=gpt-oss-120b2026.0587.23
VanillaBackbone=gpt-oss-120b2026.0587.15
CDModel Size=14B, Backbone=Qwen 2.5 14B, Algorithm=Constrained Decoding (XGrammar)2026.0286.43
TreeBackbone=gpt-oss-120b2026.0586.35
StarBackbone=gpt-oss-120b2026.0585.25
DCCDModel Size=3B, Backbone=Qwen 2.5 3B, Algorithm=Draft Conditioned Constrained Decoding2026.0284.53
8-shot, MetaBackbone=Llama-3.1-8B, Evaluation Protocol=8-shot2026.0584.5
p*LLM Family=OLMo 2 (1B and 13B)2026.0484.3
DCCDModel Size=8B, Backbone=Llama 3.1 8B, Algorithm=Draft Conditioned Constrained Decoding2026.0283.02
GRPOModel=Qwen2.5-3B, Training Data=MATH, Inference template=chat inference template, Decoding strategy=Greedy2025.0582.6
confidence betLLM Family=Qwen 3 (1.7B and 14B), lambda=0.12026.0482.4
p*LLM Family=Qwen 3 (1.7B and 14B)2026.0482.4
CFModel Size=7B, Backbone=Qwen 2.5 7B, Algorithm=Constrained Few Shot2026.0282.26
confidence betLLM Family=Qwen 3 (1.7B and 14B), lambda=0.22026.0482.1
GRPO-PVModel=Qwen2.5-3B, Training Data=MATH, Inference template=chat inference template, Decoding strategy=Greedy2025.0582
Dual KADLLM Family=Qwen 3 (1.7B and 14B), lambda=0.42026.0481.7
CDModel Size=7B, Backbone=Qwen 2.5 7B, Algorithm=Constrained Decoding (XGrammar)2026.0281.58
Dual KADLLM Family=Qwen 3 (1.7B and 14B), lambda=0.32026.0481.3
CDModel Size=8B, Backbone=Llama 3.1 8B, Algorithm=Constrained Decoding (XGrammar)2026.0280.89
Imp. rewardLLM Family=Qwen 3 (1.7B and 14B)2026.0480.7
CPModel Size=7B, Backbone=Qwen 2.5 7B, Algorithm=Constrained Prompting2026.0280.06
χ2-DRTOBackbone=Llama3-8B2026.0379.9
NudgingLLM Family=Qwen 3 (1.7B and 14B), lambda=0.42026.0479.5
INTUITORModel=Qwen2.5-3B, Training Data=MATH, Inference template=chat inference template, Decoding strategy=Greedy2025.0579.2
KL-DRTOBackbone=Llama3-8B2026.0378.5
confidence betLLM Family=Qwen 3 (1.7B and 14B), lambda=02026.0478.3
NudgingLLM Family=Qwen 3 (1.7B and 14B), lambda=0.32026.0478.1
DistiLLM-2alpha=-5, Base Model=Qwen2.5-1.5B-Instruct2025.1077.4
DistiLLM-2alpha=-1, Base Model=Qwen2.5-1.5B-Instruct2025.1076.9
CPModel Size=8B, Backbone=Llama 3.1 8B, Algorithm=Constrained Prompting2026.0276.8
pLLM Family=Qwen 3 (1.7B and 14B)2026.0475.5
q*LLM Family=Qwen 3 (1.7B and 14B)2026.0475.3
GRPOModel=Qwen2.5-1.5B, Training Data=MATH, Inference template=chat inference template, Decoding strategy=Greedy2025.0574.7
INTUITOR-CodeModel=Qwen2.5-3B, Training Data=Codeforces, Inference template=chat inference template, Decoding strategy=Greedy2025.0574.3
StudentModel=Qwen2.5-1.5B-Instruct, Role=Student2025.1074.3
DCCDModel Size=1.5B, Backbone=Qwen 2.5 1.5B, Algorithm=Draft Conditioned Constrained Decoding2026.0273.92
DPOBackbone=Llama3-8B2026.0373.9
CDModel Size=3B, Backbone=Qwen 2.5 3B, Algorithm=Constrained Decoding (XGrammar)2026.0273.24
RTOBackbone=Llama3-8B2026.0373.2
PPOBackbone=Llama3-8B2026.0373.2
GRPOBackbone=Llama3-8B2026.0372.7
Dual KADLLM Family=OLMo 2 (1B and 13B), lambda=0.42026.0472.3
CFModel Size=3B, Backbone=Qwen 2.5 3B, Algorithm=Constrained Few Shot2026.0271.8
confidence betLLM Family=OLMo 2 (1B and 13B), lambda=0.22026.0471.7
INTUITORModel=Qwen2.5-1.5B, Training Data=MATH, Inference template=chat inference template, Decoding strategy=Greedy2025.0571.1
confidence betLLM Family=OLMo 2 (1B and 13B), lambda=02026.0471.1
confidence betLLM Family=OLMo 2 (1B and 13B), lambda=0.12026.0470.5
CFModel Size=8B, Backbone=Llama 3.1 8B, Algorithm=Constrained Few Shot2026.0270.2
Dual KADLLM Family=OLMo 2 (1B and 13B), lambda=0.32026.0469.5
BaseModel=Qwen2.5-3B, Training Data=-, Inference template=chat inference template, Decoding strategy=Greedy2025.0567.3
LoRA + GASTBackbone=LLaMA3-8B2026.0366.4
0-shot CoTBackbone=Llama-3.1-8B, Evaluation Protocol=0-shot2026.0564.5
LoRA + ISTBackbone=LLaMA3-8B2026.0362.8
q*LLM Family=OLMo 2 (1B and 13B)2026.0462.5
NudgingLLM Family=OLMo 2 (1B and 13B), lambda=0.42026.0461.9
SFTBackbone=Llama3-8B2026.0361.4
LoRABackbone=LLaMA3-8B2026.0361
MDM + P2-selfModel=MDM, Parameters=1.1B, Sampling Strategy=P2-self2025.0260.9
NudgingLLM Family=OLMo 2 (1B and 13B), lambda=0.32026.0460.3
DataAgent_RLBackbone=LLaMA-DCLM, Reweighting Space=Data sources2025.0759.24
Naive TrainingBackbone=LLaMA-DCLM, Reweighting Space=Data sources2025.0759.21
CPModel Size=3B, Backbone=Qwen 2.5 3B, Algorithm=Constrained Prompting2026.0259.14
Naive TrainingBackbone=LLaMA-Nemo, Reweighting Space=Data sources2025.0759.05
Naive TrainingBackbone=LLaMA-FWE, Reweighting Space=Data sources2025.0758.91
LLaMA2Model=LLaMA2, Parameters=7B2025.0258.6
MDMModel=MDM, Parameters=1.1B, Sampling Strategy=ancestral sampling2025.0258.5
Imp. rewardLLM Family=OLMo 2 (1B and 13B)2026.0458.4
LoRA-GATrainable Parameters=319.81M2026.0458.15
DataAgent_RLBackbone=LLaMA-FWE, Reweighting Space=Data sources2025.0758.07
DataAgent_SFTBackbone=LLaMA-DCLM, Reweighting Space=Data sources2025.0757.84
ChatGPTModel=ChatGPT2026.0356.4
TLoRATrainable Parameters=171.71M2026.0456.34
DataAgent_SFTBackbone=LLaMA-FWE, Reweighting Space=Data sources2025.0756.26
DBLBackbone=LLaMA-DCLM, Reweighting Space=Data sources2025.0756.22
RegMixBackbone=LLaMA-FWE, Reweighting Space=Data sources2025.0755.9
RegMixBackbone=LLaMA-DCLM, Reweighting Space=Data sources2025.0755.87
pLLM Family=OLMo 2 (1B and 13B)2026.0454.5
CorDATrainable Parameters=319.81M2026.0454.43