List your company
ResearchFormerly WizWand

Mathematical Problem Solving on MATH

97.6AccuracyDeepSeek-R1
37.716853.263468.8184.3566Jun 18, 2024Sep 30, 2024Jan 13, 2025Apr 28, 2025Aug 11, 2025Nov 24, 2025Mar 9, 2026
Updated 7mo ago

Evaluation results

MethodLinks
DeepSeek-R1Model Category=Closed-source Models2026.0197.6
LoGos(32B)Model Category=Open-source Models (32B), Parameter Count=32B2026.0196.5
Qwen2.5-32B-InstructModel Category=Open-source Models (32B), Parameter Count=32B, Model Variant=Instruct2026.0195.2
o1-miniModel Category=Closed-source Models2026.0195
DeepSeek-R1-Distill-Qwen-32BModel Category=Open-source Models (32B), Parameter Count=32B, Model Variant=Distill2026.0194.5
LoGos(7B)Model Category=Open-source Models (7B), Parameter Count=7B2026.0193.2
Qwen2.5-7B-InstructModel Category=Open-source Models (7B), Parameter Count=7B, Model Variant=Instruct2026.0192.6
Qwen2.5-32B-BaseModel Category=Open-source Models (32B), Parameter Count=32B, Model Variant=Base2026.0190.7
DeepSeek-R1-Distill-Qwen-7BModel Category=Open-source Models (7B), Parameter Count=7B, Model Variant=Distill2026.0188.2
BaseBackbone=Qwen3-4B2025.0584
Qwen3-8BEvaluation Protocol=Zero-shot, TPS=30.112025.1283.28
Qwen2.5-7B-BaseModel Category=Open-source Models (7B), Parameter Count=7B, Model Variant=Base2026.0183.2
Qwen2.5-72BModel Type=Instruct, Parameters=72B2025.0283.1
Qwen2.5-VL-72BModel Type=Instruct, Parameters=72B2025.0283
Claude3.7-SonnetModel Category=Closed-source Models2026.0179.8
Primitives-based MASModel=DeepSeek-R1-Distill Qwen-32B2026.0279.8
M2CLModel=Qwen-72B, Number of LLMs=642026.0279.7
Primitives-based MASModel=DeepSeek-R1-Distill Llama-70B2026.0279.3
SHE-LoRABackbone=Qwen3-4B2025.0578.86
LatentMASModel=DeepSeek-R1-Distill Llama-70B2026.0278.6
LatentMASModel=DeepSeek-R1-Distill Qwen-32B2026.0278.2
Vanilla LoRABackbone=Qwen3-4B2025.0577.63
VotingModel=DeepSeek-R1-Distill Llama-70B2026.0277.2
ReviewModel=DeepSeek-R1-Distill Llama-70B2026.0275.9
PlanningModel=DeepSeek-R1-Distill Llama-70B2026.0275.6
VotingModel=DeepSeek-R1-Distill Qwen-32B2026.0274.7
PlanningModel=DeepSeek-R1-Distill Qwen-32B2026.0274.6
MacNetModel=Qwen-72B, Number of LLMs=642026.0273.9
Llama-3.1-405BModel Type=Instruct, Parameters=405B2025.0273.8
TextMASModel=DeepSeek-R1-Distill Llama-70B2026.0272.8
ReviewModel=DeepSeek-R1-Distill Qwen-32B2026.0271.3
SingleModel=DeepSeek-R1-Distill Llama-70B2026.0269.6
Qwen2-72BModel Type=Instruct, Parameters=72B2025.0269
GPTSwarmModel=Qwen-72B, Number of LLMs=642026.0268.8
Llama-3.1-70BModel Type=Instruct, Parameters=70B2025.0268
TextMASModel=DeepSeek-R1-Distill Qwen-32B2026.0268
DyLANModel=Qwen-72B, Number of LLMs=642026.0266.8
GUIDEDSAMPLINGModel=Qwen2.5-3B-Instruct, Decoding=Majority Voting2025.1064.2
Primitives-based MASModel=Qwen3-8B2026.0263.7
SingleModel=DeepSeek-R1-Distill Qwen-32B2026.0263.4
LatentMASModel=Qwen3-8B2026.0262.6
DebateModel=Qwen-72B, Number of LLMs=642026.0262.3
M2CLModel=Qwen-14B, Number of LLMs=642026.0261.9
TextMASModel=Qwen3-8B2026.0261.4
VotingModel=Qwen3-8B2026.0261.4
ReviewModel=Qwen3-8B2026.0261
SingleModel=Qwen3-8B2026.0260.8
PlanningModel=Qwen3-8B2026.0260.8
M2CLModel=Llama-70B, Number of LLMs=642026.0258
CoFiCotBase Model=GPT-3.5-Turbo2026.0357.7
Qwen-2.5-72B-Instruct-DistilledBase Model=Qwen-2.5-7B, Seed Dataset=WizardLM2025.0456.3
Qwen-2.5-32B-Instruct-DistilledBase Model=Qwen-2.5-7B, Seed Dataset=Alpaca2025.0456.26
Qwen-2.5-32B-Instruct-DistilledBase Model=Qwen-2.5-7B, Seed Dataset=Condor2025.0456
Qwen-2.5-72B-Instruct-DistilledBase Model=Qwen-2.5-7B, Seed Dataset=Alpaca2025.0455.8
Qwen-2.5-32B-Instruct-DistilledBase Model=Qwen-2.5-7B, Seed Dataset=WizardLM2025.0454.96
Qwen-2.5-72B-Instruct-DistilledBase Model=Qwen-2.5-7B, Seed Dataset=Condor2025.0454.46
REFUSIONEvaluation Protocol=Zero-shot, TPS=81.772025.1254.22
DART-Math-DSMath-7BTraining Method=SFT, Data Selection Strategy=Prop2Diff, Backbone=DeepSeekMath-7B2024.0653.6
DeepSeekMath-7B-RLTraining Method=RL, Backbone=DeepSeekMath-7B2024.0653.1
DART-Math-DSMath-7BTraining Method=SFT, Data Selection Strategy=Uniform, Backbone=DeepSeekMath-7B2024.0652.9
GRPO (RLPR)#Data=10K2026.0152.7
GRPO (RLVRR)#Data=10K2026.0152.6
Qwen3 8BCompression Ratio=dense2026.0252.57
GRPO (RM)#Data=10K2026.0152.4
Instruct#Data=n/a2026.0151.9
SFT#Data=100K2026.0151.7
GRPO (GRM)#Data=10K2026.0151.2
Repeated SamplingModel=Qwen2.5-3B-Instruct, Decoding=Majority Voting2025.1051.2
k-way SCBase Model=GPT-3.5-Turbo, k=1202026.0351.2
BoNModel=Qwen-72B, Number of LLMs=642026.0251
DPO#Data=10K2026.0151
M2CLModel=Qwen-7B, Number of LLMs=642026.0250.7
SFT#Data=10K2026.0150.6
Best-of-kBase Model=GPT-3.5-Turbo, k=1202026.0350.6
Repeated SamplingModel=Llama-3.2-3B-Instruct, Decoding=Majority Voting2025.1050.4
GRPO (BLEU)#Data=10K2026.0150.2
SDAR-4B-ChatModel=SDAR-4B-Chat, Decoding Strategy=FourierSampler2026.0150
Self-Refine + k-way SCBase Model=GPT-3.5-Turbo, k=1202026.0349.8
SDAR-4B-ChatModel=SDAR-4B-Chat, Decoding Strategy=RWS2026.0149.2
CondorBase Model=Qwen-2.5-7B, Seed Dataset=Condor2025.0448.6
SDAR-4B-ChatModel=SDAR-4B-Chat, Decoding Strategy=Vanilla2026.0148.2
CoFiCotBase Model=Llama3-8B-Instruct2026.0347.9
GRABase Model=Qwen-2.5-7B, Seed Dataset=WizardLM2025.0447.84
Dream-7B-InstructEvaluation Protocol=Zero-shot, TPS=18.992025.1246.6
Tree-of-thoughtModel=Llama-3.2-3B-Instruct, Decoding=Majority Voting2025.1045.8
Tree-of-thoughtModel=Qwen2.5-3B-Instruct, Decoding=Majority Voting2025.1045.4
MacNetModel=Llama-70B, Number of LLMs=642026.0244.2
GUIDEDSAMPLINGModel=Llama-3.2-3B-Instruct, Decoding=Majority Voting2025.1043.4
GRABase Model=Qwen-2.5-7B, Seed Dataset=Condor2025.0442.82
Best-of-kBase Model=Llama3-8B-Instruct, k=1202026.0341.4
WavefrontDiffusionModel=LLaDA-8B-Instruct, Denoising steps (T)=10242025.1141.04
SDAR-1.7B-ChatModel=SDAR-1.7B-Chat, Decoding Strategy=RWS2026.0141
Truncated-BlockDiffusion (TBD)Model=LLaDA-8B-Instruct, Denoising steps (T)=10242025.1140.84
GRPO (Random)#Data=10K2026.0140.7
BlockDiffusionModel=LLaDA-8B-Instruct, Denoising steps (T)=10242025.1140.62
k-way SCBase Model=Llama3-8B-Instruct, k=1202026.0340.6
Self-Refine + k-way SCBase Model=Llama3-8B-Instruct, k=1202026.0340.4
VanillaModel=Dream-Base, Block Size (B0)=322025.0940.2
VanillaModel=Dream-Base, Block Size (B0)=642025.0940.1
Running Confidence Remasking (RCR)Model=LLaDA-8B-Instruct, Denoising steps (T)=10242025.1140.02