Table Question Answering2015en

WikiTableQuestions

Question answering over Wikipedia tables requiring compositional reasoning

Current State of the Art

GPT-4

OpenAI

75.3

accuracy

accuracy Progress Over Time

Showing 2 breakthroughs from Apr 2020 to Jan 2024

46.054.062.070.078.0Apr 2020Jan 2024accuracyDate

Key Milestones

Apr 2020
TAPAS-large

TAPAS-large accuracy on WikiTableQuestions test set. Transfer from WikiSQL. Original TAPAS paper Table 5.

48.7
Source
Jan 2024
GPT-4Current SOTA

GPT-4 test accuracy on WikiTableQuestions. Best prior result cited in ReAcTable (VLDB 2024).

75.3
+54.6%
Source
Total Improvement
54.6%
Time Span
3y 10m
Breakthroughs
2
Current SOTA
75.3

Top Models Performance Comparison

Top 3 models ranked by accuracy

accuracy1GPT-475.3100.0%2Claude 3.5 Sonnet73.096.9%3TAPAS-large48.764.7%0%25%50%75%100%% of best
Best Score
75.3
Top Model
GPT-4
Models Compared
3
Score Range
26.6

accuracyPrimary

#ModelScorePaper / CodeDate
1
GPT-4
OpenAI		75.3
arxiv
Jan 2024
2
Claude 3.5 Sonnet
Anthropic		73
arxiv
Jan 2025
3
TAPAS-largeOpen Source
Google		48.7
TAPAS: Weakly Supervised Table Parsing via Pre-training
Apr 2020

Related Papers1

TAPAS: Weakly Supervised Table Parsing via Pre-training
Apr 2020Models: TAPAS-large
Paper

Other Table Question Answering Datasets

SQA