List your company

Technical engineering calculation on TPS-CalcBench (test)

91.4Delta KPIdeepseek-v3.1
2.68825.71948.7571.781Apr 20, 2026

Evaluation results

MethodLinks
deepseek-v3.1strategy=FT + RAG-EQ + PA-CoT, fine-tuning-dataset=DFA-TPS2026.0491.4————————
aws.claude-sonnet-4.5strategy=FT + RAG-EQ + PA-CoT, fine-tuning-dataset=DFA-TPS2026.0488.7————————
deepseek-v3.1strategy=RAG-EQ + PA-CoT2026.0487.2————————
deepseek-v3.1strategy=FT only, fine-tuning-dataset=DFA-TPS2026.0486.3————————
aws.claude-sonnet-4.5strategy=RAG-EQ + PA-CoT2026.0483.4————————
aws.claude-sonnet-4.5strategy=FT only, fine-tuning-dataset=DFA-TPS2026.0481.5————————
deepseek-v3.1strategy=Zero-shot2026.0479.22————————
gemini-3-flash-previewstrategy=FT + RAG-EQ + PA-CoT, fine-tuning-dataset=DFA-TPS2026.0472.3————————
aws.claude-sonnet-4.5strategy=Zero-shot2026.0471.75————————
gemini-3-flash-previewstrategy=RAG-EQ + PA-CoT2026.0462.1————————
gemini-3-flash-previewstrategy=FT only, fine-tuning-dataset=DFA-TPS2026.0454.8————————
gemini-3-flash-previewstrategy=Zero-shot2026.0442————————
PA-CoTBackbone Model=gemini-3-flash-preview, Strategy=PA-CoT2026.0410.81.141.451.521.251.481.221.411.54
PA-CoTBackbone Model=aws.claude-sonnet-4.5, Strategy=PA-CoT2026.047.21.641.81.781.581.741.611.721.81
PA-CoTBackbone Model=deepseek-v3.1, Strategy=PA-CoT2026.046.11.731.831.81.681.791.681.781.83
BaseBackbone Model=aws.claude-sonnet-4.5, Strategy=Base2026.04—1.621.781.711.551.410.921.331.74
BaseBackbone Model=deepseek-v3.1, Strategy=Base2026.04—1.711.821.741.661.481.031.411.77
BaseBackbone Model=gemini-3-flash-preview, Strategy=Base2026.04—1.11.421.381.211.050.610.881.35