Spreadsheet Understanding on SpreadsheetLLM benchmark (298 queries)
98AccuracyGPT-5
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GPT-5Retrieval Technique=BRTR2026.03 | 98 | 156,306 | 273.36 | |
| Claude Sonnet 4.5Retrieval Technique=BRTR2026.03 | 97 | 166,117 | 280.59 | |
| Claude Sonnet 4.5Retrieval Technique=SpreadsheetLLM2026.03 | 91 | — | — | |
| GPT-5Retrieval Technique=SpreadsheetLLM2026.03 | 90 | — | — | |
| Gemini 2.5 ProRetrieval Technique=SpreadsheetLLM2026.03 | 89 | — | — | |
| GPT-5Retrieval Technique=FRTR2026.03 | 87 | — | — | |
| Gemini 2.5 ProRetrieval Technique=FRTR2026.03 | 85 | — | — | |
| Claude Sonnet 4.5Retrieval Technique=FRTR2026.03 | 84 | — | — | |
| Gemini 2.5 ProRetrieval Technique=BRTR2026.03 | 79 | 27,675 | 223.79 | |
| GPT-4oRetrieval Technique=SpreadsheetLLM2026.03 | 78 | — | — | |
| GPT-4oRetrieval Technique=FRTR2026.03 | 77 | — | — | |
| LLaMA 4 Maverick 17BRetrieval Technique=SpreadsheetLLM2026.03 | 73 | — | — | |
| GPT-5Retrieval Technique=Naïve Full-Context2026.03 | 68 | — | — | |
| Gemini 2.5 ProRetrieval Technique=Naïve Full-Context2026.03 | 67 | — | — | |
| GPT-4oRetrieval Technique=BRTR2026.03 | 56 | 256,523 | 315.93 | |
| Claude Sonnet 4.5Retrieval Technique=Naïve Full-Context2026.03 | 55 | — | — | |
| LLaMA 4 Maverick 17BRetrieval Technique=FRTR2026.03 | 53 | — | — | |
| LLaMA 4 Maverick 17BRetrieval Technique=Naïve Full-Context2026.03 | 41 | — | — | |
| GPT-4oRetrieval Technique=Naïve Full-Context2026.03 | 35 | — | — | |
| LLaMA 4 Maverick 17BRetrieval Technique=BRTR2026.03 | 30 | 18,608 | 408.73 |