Spreadsheet Understanding and Manipulation on SpreadsheetBench Soft Restriction
75.56Cell-Level AccuracyHuman Performance
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Human Performance2026.04 | 75.56 | 65 | 71.33 | |
| SpreadsheetAgentBackbone=Qwen3-Coder-480B2026.04 | 45.63 | 35.33 | 41.67 | |
| SpreadsheetAgentBackbone=GPT-OSS-120B2026.04 | 41.3 | 33.14 | 38.16 | |
| TreeThinkerBackbone=Qwen3-Coder-480B2026.04 | 38.86 | 32 | 36.22 | |
| ChatGPT Agent2026.04 | 38.27 | 30.48 | 35.27 | |
| SpreadsheetAgentBackbone=GPT-OSS-20B2026.04 | 34.46 | 28.49 | 32.16 | |
| SpreadsheetAgentBackbone=Qwen3-235B2026.04 | 33.1 | 28.77 | 31.43 | |
| TreeThinkerBackbone=GPT-OSS-120B2026.04 | 32.14 | 29.82 | 31.25 | |
| GPT-OSS-120BBackbone=GPT-OSS-120B2026.04 | 30.78 | 27.64 | 29.57 | |
| Qwen3-Coder-480BBackbone=Qwen3-Coder-480B2026.04 | 30.36 | 31.05 | 30.63 | |
| GPT-OSS-20BBackbone=GPT-OSS-20B2026.04 | 25.55 | 24.79 | 25.26 | |
| TreeThinkerBackbone=GPT-OSS-20B2026.04 | 23.95 | 26.12 | 24.78 | |
| Copilot in Excel2026.04 | 23.33 | 15 | 20 | |
| TreeThinkerBackbone=Qwen3-235B2026.04 | 22.99 | 26.5 | 24.34 | |
| SpreadsheetAgentBackbone=Qwen3-30B2026.04 | 20.62 | 25.17 | 22.37 | |
| Qwen3-235BBackbone=Qwen3-235B2026.04 | 20.5 | 25.83 | 22.55 | |
| SheetCopilot2026.04 | 16.67 | 10 | 14 | |
| TreeThinkerBackbone=Qwen3-30B2026.04 | 14.38 | 22.22 | 17.4 | |
| Qwen3-30BBackbone=Qwen3-30B2026.04 | 13.61 | 20.89 | 16.41 | |
| GPT-4oBackbone=GPT-4o2026.04 | 13.49 | 22.51 | 16.96 | |
| Mixtral-8x7BBackbone=Mixtral-8x7B2026.04 | 3.39 | 4.67 | 3.88 | |
| GPT-3.5Backbone=GPT-3.52026.04 | 3.33 | 13.11 | 7.09 | |
| BinderBackbone=GPT-3.52026.04 | 1.58 | 0.05 | 1.17 | |
| Llama-3-70BBackbone=Llama-3-70B2026.04 | 1.13 | 7.9 | 3.74 |