List your company
ResearchFormerly WizWand

Table Question Answering on WikiTQ (dev)

87.8Denotation AccCRAFT
9.675229.957650.2470.5224Jan 16, 2022Oct 9, 2022Jul 3, 2023Mar 26, 2024Dec 18, 2024Sep 11, 2025Jun 5, 2026
Updated 4mo ago

Evaluation results

MethodLinks
CRAFTBackbone LLM=GPT-5-mini2026.0687.8——
CRAFT w/ Chain-of-Table ExtractorBackbone LLM=GPT-5-mini2026.0687——
Table-CriticBackbone LLM=GPT-5-mini2026.0686.7——
Chain-of-TableBackbone LLM=GPT-5-mini2026.0686.2——
Critic-CoTBackbone LLM=GPT-5-mini2026.0686.1——
Few-ShotBackbone LLM=GPT-5-mini2026.0685.7——
End-to-EndBackbone LLM=GPT-5-mini2026.0685.2——
DaterBackbone LLM=GPT-5-mini2026.0684.9——
BinderBackbone LLM=GPT-5-mini2026.0684.2——
CRAFTBackbone LLM=Average2026.0682.4——
CRAFTBackbone LLM=Llama3.3-70B2026.0681.4——
CRAFT w/ Chain-of-Table ExtractorBackbone LLM=Average2026.0680.7——
CRAFTBackbone LLM=Qwen2.5-72B2026.0680.4——
CRAFT w/ Chain-of-Table ExtractorBackbone LLM=Llama3.3-70B2026.0679.9——
CRAFTBackbone LLM=Deepseek-R1-14B2026.0679.9——
CRAFT w/ Chain-of-Table ExtractorBackbone LLM=Qwen2.5-72B2026.0678.7——
Table-CriticBackbone LLM=Qwen2.5-72B2026.0677.7——
Table-CriticBackbone LLM=Average2026.0677.7——
CRAFT w/ Chain-of-Table ExtractorBackbone LLM=Deepseek-R1-14B2026.0677.2——
Table-CriticBackbone LLM=Deepseek-R1-14B2026.0676.3——
Critic-CoTBackbone LLM=Deepseek-R1-14B2026.0675.2——
Chain-of-TableBackbone LLM=Deepseek-R1-14B2026.0675.1——
Critic-CoTBackbone LLM=Average2026.0674.3——
Chain-of-TableBackbone LLM=Average2026.0673——
DaterBackbone LLM=Deepseek-R1-14B2026.0671.6——
DaterBackbone LLM=Average2026.0670.9——
AlterBackbone LLM=Qwen2.5-72B2026.0670.9——
Table-CriticBackbone LLM=Llama3.3-70B2026.0670.1——
Few-ShotBackbone LLM=Deepseek-R1-14B2026.0670——
Few-ShotBackbone LLM=Average2026.0669.6——
BinderBackbone LLM=Average2026.0669——
Critic-CoTBackbone LLM=Qwen2.5-72B2026.0669——
AlterBackbone LLM=Average2026.0668.7——
BinderBackbone LLM=Deepseek-R1-14B2026.0668.5——
Chain-of-TableBackbone LLM=Qwen2.5-72B2026.0668.3——
AlterBackbone LLM=Llama3.3-70B2026.0667.8——
AlterBackbone LLM=Deepseek-R1-14B2026.0667.4——
Critic-CoTBackbone LLM=Llama3.3-70B2026.0666.8——
BinderBackbone LLM=Qwen2.5-72B2026.0665.5——
DaterBackbone LLM=Qwen2.5-72B2026.0663.8——
DaterBackbone LLM=Llama3.3-70B2026.0663.3——
Chain-of-TableBackbone LLM=Llama3.3-70B2026.0662.1——
End-to-EndBackbone LLM=Average2026.0662——
Few-ShotBackbone LLM=Llama3.3-70B2026.0662——
Few-ShotBackbone LLM=Qwen2.5-72B2026.0660.6——
REASTAPBackbone=BART-Large2022.1058.3——
TAPEX2022.1058——
BinderBackbone LLM=Llama3.3-70B2026.0657.6——
End-to-EndBackbone LLM=Qwen2.5-72B2026.0657.3——
End-to-EndBackbone LLM=Deepseek-R1-14B2026.0654.3——
TaBERT2022.1053——
GraPPa2022.1051.9——
TableFormer2022.1051.3——
End-to-EndBackbone LLM=Llama3.3-70B2026.0651.1——
T5-3BMode=finetune2022.0150.65——
TaPas2022.1049.9——
LatentAlignment2022.1043.7——
MeRL2022.1043.2——
IterativeSearch2022.1043.1——
MAPO2022.1042.7——
MacroGrammer2022.1040.6——
BARTBackbone=BART-Large2022.1037.1——
GPT-3 175BFew-shot selection strategy=select2022.0132——
GPT-3 175BFew-shot selection strategy=random2022.0129.33——
Codex 175BFew-shot selection strategy=select2022.0126.21——
Codex 175BFew-shot selection strategy=random2022.0120.46——
T0 3BMode=zero-shot2022.0112.68——
Codex BinderFinetuning status=No2023.02——65
Codex QA*Finetuning status=No, Modeling approach=end-to-end QA2023.02——50.5
Codex SQLFinetuning status=No2023.02——60.2
DaterBase LLM=GPT-3.52025.06—58.4—
Dater + CoREBase LLM=GPT-3.52025.06—63.5—
EP + MLBase Model=code-davinci-0022023.02——52.7
GreedyBase Model=code-davinci-0022023.02——49.6
LEVERBase Model=code-davinci-0022023.02——64.6
StructGPTBase LLM=GPT-3.52025.06—64.4—
StructGPT + CoREBase LLM=GPT-3.52025.06—66.1—
TaCubeFinetuning status=Yes2023.02——61.1
TaPEX*Finetuning status=Yes, Modeling approach=end-to-end QA2023.02——60.4