List your company
ResearchFormerly WizWand

Mathematical Reasoning on ASDiv (Accuracy)

0.955AccuracyQwen2.5-14B-Instruct-1M
0.8510.8780.9050.932Feb 15, 2024Jul 3, 2024Nov 20, 2024Apr 8, 2025Aug 26, 2025Jan 12, 2026Jun 1, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Qwen2.5-14B-Instruct-1MSize=14B, Type=Instruct, Context Length=1M2025.080.955
Ori-SFTModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=Ori-SFT, Evaluation Protocol=Zero-shot2025.060.953
AbstRaLModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=AbstRaL, Evaluation Protocol=Zero-shot2025.060.953
BERT-Judgeclean_name=BERT-as-a-Judge2026.040.953
NPG-Muse-8BBackbone=Qwen3-8B-Base2025.080.952
CoAModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=CoA, Evaluation Protocol=Zero-shot2025.060.951
CoT-RLModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=CoT-RL, Evaluation Protocol=Zero-shot2025.060.949
Qwen2.5-7B-Instruct-1MSize=7B, Type=Instruct, Context Length=1M2025.080.947
Qwen3-14B-BaseSize=14B, Type=Base2025.080.944
eMoTModels=Qwen-32B2026.060.944
GPT-42024.050.943
NPG-Muse-7BBackbone=Qwen2.5-7B-Instruct-1M2025.080.939
JiuZhang3.0-8x7BModel Scale=8x7B2024.050.931
LEMMABackbone=Qwen2-Math-7B, # Samples=88.90k2025.030.93
BoTModels=GPT-42026.060.928
GPT-4Protocol=PAL, Setting=Tool manipulation2024.050.926
JiuZhang3.0-7BModel Scale=7B, Backbone=Mistral-7B2024.050.926
JiuZhang3.0-8BModel Scale=8B, Backbone=LLaMA-3-8B2024.050.926
GPT-4 (Code Interpreter)Decoding Strategy=Greedy2024.020.926
Qwen2-7B-SFT + Full-Step-DPOBackbone=Qwen2-7B, Training Strategy=Full-Step-DPO2025.020.924
RefAug-90kBackbone=Qwen2-Math-7B, # Samples=89.92k2025.030.924
GPTAugBackbone=Qwen2-Math-7B, # Samples=88.62k2025.030.923
MAmmoTH2-8x7B-PlusModel Scale=8x7B2024.050.922
RefAugBackbone=Qwen2-Math-7B, # Samples=29.94k2025.030.92
DeepSeekMath-7B-RLModel Scale=7B2024.050.918
Qwen2-7B-SFT + Step-DPOBackbone=Qwen2-7B, Training Strategy=Step-DPO2025.020.918
COMT+CCRLBackbone=QWEN3-4B2026.010.917
Qwen2-7B-SFTBackbone=Qwen2-7B, Training Strategy=SFT2025.020.916
DeepSeekMath-Base-SFT + Full-Step-DPOBackbone=DeepSeekMath-Base, Training Strategy=Full-Step-DPO2025.020.912
DeepSeekMath-Base-SFTBackbone=DeepSeekMath-Base, Training Strategy=SFT2025.020.91
FEW-SHOTBackbone=QWEN2.5-7B2026.010.909
COMTBackbone=QWEN3-8B2026.010.909
QWEN2.5-7BModel=QWEN2.5-7B, Method=FEW-SHOT2026.010.909
Base LLMBackbone=Qwen2.5-7B, Mode=Few-shot2026.010.909
Base LLMBackbone=Qwen2.5-7B, Mode=Few-shot2026.010.909
CoT-SFTBackbone=Qwen2-1.5B, Teacher Source=Qwen2026.010.908
FSLRBackbone=Qwen2-1.5B, Teacher Source=Qwen2026.010.908
DeepSeekMath-Base-SFT + Step-DPOBackbone=DeepSeekMath-Base, Training Strategy=Step-DPO2025.020.907
RFTBackbone=Qwen2-Math-7B, # Samples=86.52k2025.030.907
COMT+CCRLModel=QWEN3-8B, Method=COMT+CCRL2026.010.907
FSLRBackbone=Qwen2-1.5B, Teacher Source=Self2026.010.907
DeepSeekMath-Base-SFT + DPOBackbone=DeepSeekMath-Base, Training Strategy=DPO2025.020.906
FSLRBackbone=Qwen2.5-7B, Data Source=LLaMA, Training Set=SVAMP2026.010.905
MetaMathBackbone=Qwen2-Math-7B, # Samples=394.99k2025.030.904
FSLRBackbone=Qwen3-4B, Data Source=Self, Training Set=SVAMP2026.010.904
MAmmoTH2-7B-PlusModel Scale=7B2024.050.903
FSLRBackbone=Qwen3-4B, Data Source=Qwen, Training Set=SVAMP2026.010.902
DeepSeekMath-7B-InstructModel Scale=7B2024.050.901
MAmmoTH2-8B-PlusModel Scale=8B2024.050.899
COMT+CCRLModel=QWEN3-4B, Method=COMT+CCRL2026.010.899
FSLRBackbone=Qwen3-4B, Data Source=LLaMA, Training Set=SVAMP2026.010.899
CoT-SFTBackbone=Qwen3-4B, Data Source=Qwen, Training Set=SVAMP2026.010.899
COMTBackbone=QWEN3-4B2026.010.898
FEW-SHOTBackbone=QWEN3-8B2026.010.897
QWEN3-8BModel=QWEN3-8B, Method=FEW-SHOT2026.010.897
COMTModel=QWEN2.5-7B, Method=COMT2026.010.896
ISCBackbone=Qwen2-Math-7B, # Samples=86.78k2025.030.894
FSLRBackbone=Qwen2.5-7B, Teacher Source=Qwen2026.010.893
Regex2026.040.892
COMT+CCRLBackbone=QWEN3-8B2026.010.891
CoT-SFTBackbone=Qwen3-4B, Data Source=Self, Training Set=SVAMP2026.010.891
KPMath-DSMath-7BModel Scale=7B2024.050.889
COMT+CCRLBackbone=QWEN2.5-7B2026.010.888
COMT+CCRLModel=QWEN2.5-7B, Method=COMT+CCRL2026.010.888
COT-SFT+RLBackbone=QWEN2.5-7B2026.010.887
FSLRBackbone=Qwen2.5-7B, Teacher Source=LLaMA2026.010.886
FEW-SHOTBackbone=QWEN3-4B2026.010.884
COT-SFT+RLModel=QWEN2.5-7B, Method=COT-SFT+RL2026.010.884
QWEN3-4BModel=QWEN3-4B, Method=FEW-SHOT2026.010.884
Base LLMBackbone=Qwen2-1.5B, Mode=Few-shot2026.010.884
Base LLMBackbone=Qwen3-4B, Mode=Few-shot2026.010.884
JiuZhang3.0-7BSetting=Tool manipulation, Model Size=7B2024.050.883
SFTBackbone=Qwen2-Math-7B, # Samples=14.97k2025.030.881
FSLRBackbone=LLaMA3.1-8B, Teacher Source=Qwen2026.010.878
ChatGPT2024.050.877
COMTModel=QWEN3-8B, Method=COMT2026.010.876
JiuZhang3.0-8BSetting=Tool manipulation, Model Size=8B2024.050.875
COMTBackbone=QWEN2.5-7B2026.010.874
COMTModel=QWEN3-4B, Method=COMT2026.010.869
CoT-SFTBackbone=Qwen2.5-7B, Data Source=LLaMA, Training Set=SVAMP2026.010.869
ToRAModel Size=70B, Base Model=Llama-2, Decoding Strategy=Greedy2024.020.868
COT-SFTBackbone=QWEN2.5-7B2026.010.868
ElasticModel=LLaDA2.0-mini2026.050.8672
FSLRBackbone=LLaMA3.1-8B, Teacher Source=LLaMA2026.010.867
Qwen3-8B-BaseSize=8B, Type=Base2025.080.866
FSLRBackbone=Qwen2.5-7B, Data Source=Qwen, Training Set=SVAMP2026.010.865
Llama-3-8B-SFT + Full-Step-DPOBackbone=Llama-3-8B, Training Strategy=Full-Step-DPO2025.020.864
CoT-SFTBackbone=Qwen2-1.5B, Teacher Source=Self2026.010.862
BaselineModel=LLaDA2.0-mini2026.050.862
FSLRBackbone=Qwen2.5-7B, Data Source=Self, Training Set=SVAMP2026.010.86
FSLRBackbone=LLaMA3.1-8B, Data Source=LLaMA, Training Set=SVAMP2026.010.859
FEW-SHOTBackbone=LLAMA3.1-8B2026.010.858
LLAMA3.1-8BModel=LLAMA3.1-8B, Method=FEW-SHOT2026.010.858
Base LLMBackbone=LLaMA3.1-8B, Mode=Few-shot2026.010.858
Base LLMBackbone=LLaMA3.1-8B, Mode=Few-shot2026.010.858
COMT+CCRLBackbone=LLAMA3.1-8B2026.010.857
COT-SFTModel=QWEN3-4B, Method=COT-SFT2026.010.857
FSLRBackbone=Qwen2.5-7B, Teacher Source=Self2026.010.856
Rho-1-Math-7B-CodeSetting=Tool manipulation, Model Size=7B2024.050.855
ZERO-SHOTBackbone=QWEN2.5-MATH2026.010.855