List your company
ResearchFormerly WizWand

Mathematical Reasoning on TabMWP (accuracy)

97.61AccuracyZero-shot-EI
67.647675.426383.20590.9837Feb 15, 2024Jul 5, 2024Nov 23, 2024Apr 13, 2025Sep 1, 2025Jan 20, 2026Jun 10, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Zero-shot-EIModel=Gemini-1.5-Pro, Model Role=Teacher Model2026.0297.61
ExGRPOModel=Qwen2.5-7B-Instruct, Model Role=Student Model, Adapter=On-Policy Adapt.2026.0297.43
NPG-Muse-8BBackbone=Qwen3-8B-Base2025.0897.2
ExGRPOModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0297.1
Qwen2.5-14B-Instruct-1MSize=14B, Type=Instruct, Context Length=1M2025.0896.9
RevThinkModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0296.73
On-Policy DistillationModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0296.15
Answer AugModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0296.12
MOCA-AGENTMethod Category=Reinforcement learning, Backbone=Qwen3.6-27B2026.0696
GPT-4Protocol=PAL, Setting=Tool manipulation2024.0595.9
GPT-4 (Code Interpreter)Decoding Strategy=Greedy2024.0295.9
Question AugModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0295.67
Divide-or-ConquerModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0295.62
gemma-3-12b-itBackbone=Gemma 3, RL Type=/, Supervision=/, Environment=/2026.0595.2
Rephrase QuestionModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0295.12
Zero-shotModel=Gemini-1.5-Pro, Model Role=Teacher Model2026.0295.1
Distill Step-by-StepModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0294.76
CREATORMethod Category=Tool-augmented / agent-based2026.0694.7
SKDModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0294.17
Ori-SFTModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=Ori-SFT, Evaluation Protocol=Zero-shot2025.0693.9
AbstRaLModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=AbstRaL, Evaluation Protocol=Zero-shot2025.0693.9
Zero-shotModel=Qwen2.5-7B-Instruct, Model Role=Student Model2026.0293.88
Qwen2.5-14B-InstructBackbone=Qwen 2.5, RL Type=/, Supervision=/, Environment=/2026.0593.7
ChameleonMethod Category=Tool-augmented / agent-based2026.0693.28
Qwen3-14B-BaseSize=14B, Type=Base2025.0893.2
CoAModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=CoA, Evaluation Protocol=Zero-shot2025.0692.9
PoT+DocMethod Category=Tool-augmented / agent-based2026.0692.69
SimpleSEBackbone=Qwen 2.5, RL Type=Offline, Supervision=No, Environment=No2026.0592.3
Qwen2.5-7B-InstructBackbone=Qwen 2.5, RL Type=/, Supervision=/, Environment=/2026.0591.9
ExGRPOModel=Gemma-7B-it, Model Role=Student Model, Adapter=On-Policy Adapt.2026.0291.23
CoTMethod Category=Prompting / in-context learning2026.0690.81
GPT-42024.0590.8
ExGRPOModel=Gemma-7B-it, Model Role=Student Model2026.0290.55
NPG-Muse-7BBackbone=Qwen2.5-7B-Instruct-1M2025.0889.9
PoTMethod Category=Prompting / in-context learning2026.0689.49
CoT-RLModel=Qwen2.5-Math-7B-Instruct, GSM Train Method=CoT-RL, Evaluation Protocol=Zero-shot2025.0689.2
CRITICMethod Category=Tool-augmented / agent-based2026.0689
CPMajModel=Average across models2026.0488.4
CPMaxModel=Average across models2026.0488.4
CPAgrModel=Average across models2026.0488.4
SCCoTModel=Average across models2026.0488.2
SimpleSEBackbone=Gemma 3, RL Type=Offline, Supervision=No, Environment=No2026.0587.4
CoTModel=Average across models2026.0487.1
SCPoTModel=Average across models2026.0486.3
On-Policy DistillationModel=Gemma-7B-it, Model Role=Student Model2026.0286.21
RevThinkModel=Gemma-7B-it, Model Role=Student Model2026.0284.71
JiuZhang3.0Parameters=8x7B2024.0584.7
gemma-3-4b-itBackbone=Gemma 3, RL Type=/, Supervision=/, Environment=/2026.0584.5
Qwen3-8B-BaseSize=8B, Type=Base2025.0884
ChatGPT2024.0582
Qwen2.5-7B (Base) + RL w/ SIE-100%Base Model=Qwen2.5-7B (Base), Learning/Inference Strategy=RL w/ SIE-100%2025.0981.8
PoTModel=Average across models2026.0481.6
Qwen-1.5Parameters=110B2024.0580.5
ChatGPTProtocol=PAL, Setting=Tool manipulation2024.0579.9
JiuZhang3.0-8BSetting=Tool manipulation, Model Size=8B2024.0579.9
JiuZhang3.0Parameters=8B2024.0579.2
Base + AZR-CoderBackbone=Qwen 2.5, RL Type=Online, Supervision=No, Environment=Yes2026.0578.5
Qwen2.5-7B-Instruct-1MSize=7B, Type=Instruct, Context Length=1M2025.0878.4
Llama3.1-8B-Instruct + RL w/ SIE-100%Base Model=Llama3.1-8B-Instruct, Learning/Inference Strategy=RL w/ SIE-100%2025.0977
JiuZhang3.0-7BSetting=Tool manipulation, Model Size=7B2024.0575.6
Divide-or-ConquerModel=Gemma-7B-it, Model Role=Student Model2026.0275.18
MAmmoTH2Parameters=8B, Variant=Plus2024.0575.1
JiuZhang3.0Parameters=7B2024.0574.8
COMTModel=QWEN3-8B, Method=COMT2026.0174.3
OpenMath-CodeLlamaModel Size=70B, Base Model=CodeLlama, Decoding Strategy=Greedy2024.0274.2
COMT+CCRLBackbone=QWEN3-8B2026.0174.2
ToRAModel Size=70B, Base Model=Llama-2, Decoding Strategy=Greedy2024.0274
FEW-SHOTBackbone=QWEN3-8B2026.0173.9
QWEN3-8BModel=QWEN3-8B, Method=FEW-SHOT2026.0173.9
COMTBackbone=QWEN3-8B2026.0173.6
COMT+CCRLModel=QWEN3-8B, Method=COMT+CCRL2026.0172.8
CoT-SFTBackbone=Qwen3-4B, Data Source=Self, Training Set=SVAMP2026.0171.5
OpenMath-Llama2Model Size=70B, Base Model=Llama-2, Decoding Strategy=Greedy2024.0270.8
DeepSeekMathParameters=7B, Variant=Instruct2024.0570.5
ToRAModel Size=34B, Base Model=CodeLlama, Decoding Strategy=Greedy2024.0270.5
FEW-SHOTBackbone=QWEN2.5-7B2026.0170.4
QWEN2.5-7BModel=QWEN2.5-7B, Method=FEW-SHOT2026.0170.4
Base LLMBackbone=Qwen2.5-7B, Mode=Few-shot2026.0170.4
Base LLMBackbone=Qwen2.5-7B, Mode=Few-shot2026.0170.4
FEW-SHOTBackbone=QWEN3-4B2026.0170.3
QWEN3-4BModel=QWEN3-4B, Method=FEW-SHOT2026.0170.3
Base LLMBackbone=Qwen2-1.5B, Mode=Few-shot2026.0170.3
Base LLMBackbone=Qwen3-4B, Mode=Few-shot2026.0170.3
FSLRBackbone=Qwen3-4B, Data Source=Self, Training Set=SVAMP2026.0170.2
Rho-1-Math-7B-CodeSetting=Tool manipulation, Model Size=7B2024.0570.1
OpenMath-Mistral-7BSetting=Tool manipulation, Model Size=7B, Backbone=Mistral2024.0570
OpenMath-Mistral-7BModel Size=7B, Base Model=Mistral, Decoding Strategy=Greedy2024.0270
CoT-SFTBackbone=LLaMA3.1-8B, Teacher Source=Qwen2026.0170
ZERO-SHOTBackbone=DEEPSEEK-MATH2026.0169.9
DeepSeek-MathModel=DeepSeek-Math, Method=ZERO-SHOT2026.0169.9
DeepSeek-MathMode=Zero-shot2026.0169.9
DeepSeek-MathBackbone=DeepSeek-Math-7B, Mode=Zero-shot2026.0169.9
DeepSeekMathParameters=7B2024.0569.8
COT-SFTModel=QWEN3-8B, Method=COT-SFT2026.0169.7
COMT+CCRLBackbone=QWEN3-4B2026.0169.6
FSLRBackbone=Qwen2-1.5B, Teacher Source=Self2026.0169.6
COMT+CCRLModel=QWEN3-4B, Method=COMT+CCRL2026.0169.5
Llama3.1-8B-Instruct + CoTBase Model=Llama3.1-8B-Instruct, Learning/Inference Strategy=CoT2025.0969.5
FSLRBackbone=Qwen2-1.5B, Teacher Source=Qwen2026.0169.1
Base + AZRBackbone=Qwen 2.5, RL Type=Online, Supervision=No, Environment=Yes2026.0568.8