List your company
ResearchFormerly WizWand

Optimization Modeling on OptMATH

87.5Accuracy Rate (AR)TriVAL
-0.79622.12745.0567.973Feb 2, 2026Feb 23, 2026Mar 17, 2026Apr 8, 2026Apr 30, 2026May 22, 2026Jun 13, 2026
Updated 4mo ago

Evaluation results

MethodLinks
TriVALModel=DeepSeek-V3.2, Method category=Prompt-based2026.0587.5
TriVALModel=GPT-5.1, Method category=Prompt-based2026.0587.5
OR-LLM-AgentModel=DeepSeek-V3.2, Method category=Prompt-based2026.0585.9
OR-LLM-AgentModel=GPT-5.1, Method category=Prompt-based2026.0585.2
OptiTreeModel=GPT-5.1, Method category=Prompt-based2026.0583.6
OptiTreeModel=DeepSeek-V3.2, Method category=Prompt-based2026.0574.2
SIRLMethod category=Learning-based2026.0571.1
DCM-AgentBackbone=GPT5.12026.0455.42
OptiTreeBackbone=GPT5.12026.0453.61
DCM-AgentBackbone=Deepseek-V3.22026.0452.73
R2C with reflectionReflection mechanism=Enabled, Maximum reflection iterations=32026.0251.8
AF-MCTSBackbone=GPT5.12026.0451.2
OptiTreeBackbone=Deepseek-V3.22026.0447.59
Grok-42026.0247
Grok42026.0247
AutoFormulationModel=GPT-5.1, Method category=Prompt-based2026.0546.9
DCM-AgentBackbone=Qwen3-235B2026.0446.39
GPT-52026.0245.8
GPT52026.0245.8
OptiMUSBackbone=GPT5.12026.0445.78
R2CReflection mechanism=Disabled2026.0244.6
R2CBackbone=deepseek-v3.22026.0244.6
Standard PromptBackbone=deepseek-v3.22026.0244
OpenAI-o3Method Category=Model-based Methods2026.0644
AF-MCTSBackbone=Deepseek-V3.22026.0443.98
DeepSeek-V3.1Method Category=Model-based Methods2026.0643.9
BaselineBackbone=GPT5.12026.0443.38
OptiTreeBackbone=Qwen3-235B2026.0441.57
AutoFormulationModel=DeepSeek-V3.2, Method category=Prompt-based2026.0540.6
DeepSeek-R1Method Category=Model-based Methods2026.0640.4
AF-MCTSBackbone=Qwen3-235B2026.0440.36
OptiMUSBackbone=Deepseek-V3.22026.0439.76
OptiMUSBackbone=Qwen3-235B2026.0437.95
BaselineBackbone=Deepseek-V3.22026.0436.75
OptMATHBackbone=Qwen2.5-32B2026.0234.7
DCM-AgentBackbone=Qwen3-30B2026.0432.53
StarORMethod Category=Test-Time Methods, Base model=Qwen3-4B-Instruct-25072026.0632.5
BaselineBackbone=Qwen3-235B2026.0431.33
SIRL-Qwen2.5-7BMethod Category=Learning-based Methods2026.0630.5
OptiTreeBackbone=Qwen3-30B2026.0427.71
AF-MCTSBackbone=Qwen3-30B2026.0425.9
OptiMUSBackbone=Qwen3-30B2026.0424.7
OptMATH-Qwen2.5-7BMethod Category=Learning-based Methods2026.0624.4
LLMOPTMethod category=Learning-based2026.0521.9
DCM-AgentBackbone=Qwen3-8B2026.0421.69
Best-of-NMethod Category=Test-Time Methods, Base model=Qwen3-4B-Instruct-2507, N=162026.0621.6
BaselineBackbone=Qwen3-30B2026.0416.87
GPT-42026.0216.6
GPT-4Method Category=Model-based Methods2026.0616.6
OptiTreeBackbone=Qwen3-8B2026.0416.26
OptiMUSBackbone=Qwen3-8B2026.0415.06
ORLMMethod category=Learning-based2026.0514.1
OptiTreeMethod Category=Test-Time Methods, Base model=Qwen3-4B-Instruct-25072026.0612.7
AF-MCTSBackbone=Qwen3-8B2026.0412.65
LLMOPT-Qwen2.5-14BMethod Category=Learning-based Methods2026.0612.5
ReflexionMethod Category=Test-Time Methods, Base model=Qwen3-4B-Instruct-25072026.0611.4
AutoFormulatorMethod Category=Test-Time Methods, Base model=Qwen3-4B-Instruct-25072026.0611.4
Zero-shotMethod Category=Test-Time Methods, Base model=Qwen3-4B-Instruct-25072026.0610.8
BaselineBackbone=Qwen3-8B2026.0410.24
ORLM-LLaMA3-8BMethod Category=Learning-based Methods2026.062.6