Chart-to-table extraction on PlotQA n=529
99.1RMSF1Gemini 3 Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini 3 FlashProvider=Google, Evaluation Protocol=best-column matching, Prompt Complexity=simple2026.05 | 99.1 | |
| Claude Sonnet 4.6Provider=Anthropic, Evaluation Protocol=best-column matching, Prompt Complexity=simple2026.05 | 98.8 | |
| GPT-5.4 miniProvider=OpenAI, Evaluation Protocol=best-column matching, Prompt Complexity=simple2026.05 | 98.3 | |
| Gemini 3.1 Flash LiteProvider=Google, Evaluation Protocol=best-column matching, Prompt Complexity=simple2026.05 | 98.1 | |
| Claude Haiku 4.5Provider=Anthropic, Evaluation Protocol=best-column matching, Prompt Complexity=simple2026.05 | 96.3 | |
| DePlot (published)Evaluation Protocol=best-column matching2026.05 | 94.2 | |
| GPT-5.4 nanoProvider=OpenAI, Evaluation Protocol=best-column matching, Prompt Complexity=simple2026.05 | 86.3 |