Spreadsheet Understanding and Manipulation on SpreadsheetBench Hard Restriction
66.67Cell-Level AccuracyHuman Performance
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Human Performance2026.04 | 66.67 | 55 | 62 | |
| SpreadsheetAgentBackbone=Qwen3-Coder-480B2026.04 | 36.9 | 31.05 | 34.65 | |
| SpreadsheetAgentBackbone=GPT-OSS-120B2026.04 | 32.8 | 29.34 | 31.47 | |
| TreeThinkerBackbone=Qwen3-Coder-480B2026.04 | 31.91 | 27.92 | 30.37 | |
| SpreadsheetAgentBackbone=GPT-OSS-20B2026.04 | 27.81 | 25.64 | 29.25 | |
| SpreadsheetAgentBackbone=Qwen3-235B2026.04 | 25.31 | 25.07 | 25.22 | |
| GPT-OSS-120BBackbone=GPT-OSS-120B2026.04 | 24.96 | 23.93 | 24.56 | |
| TreeThinkerBackbone=GPT-OSS-120B2026.04 | 24.42 | 26.5 | 25.22 | |
| Qwen3-Coder-480BBackbone=Qwen3-Coder-480B2026.04 | 22.82 | 27.07 | 24.45 | |
| GPT-OSS-20BBackbone=GPT-OSS-20B2026.04 | 19.43 | 21.37 | 20.18 | |
| TreeThinkerBackbone=GPT-OSS-20B2026.04 | 18.54 | 22.51 | 20.07 | |
| TreeThinkerBackbone=Qwen3-235B2026.04 | 17.29 | 23.65 | 19.74 | |
| SpreadsheetAgentBackbone=Qwen3-30B2026.04 | 16.04 | 22.22 | 18.42 | |
| Qwen3-235BBackbone=Qwen3-235B2026.04 | 15.33 | 22.51 | 18.09 | |
| GPT-4oBackbone=GPT-4o2026.04 | 10.52 | 17.66 | 13.27 | |
| TreeThinkerBackbone=Qwen3-30B2026.04 | 10.16 | 19.09 | 13.6 | |
| Qwen3-30BBackbone=Qwen3-30B2026.04 | 9.45 | 18.23 | 12.83 | |
| GPT-3.5Backbone=GPT-3.52026.04 | 2.5 | 9.97 | 5.37 | |
| Mixtral-8x7BBackbone=Mixtral-8x7B2026.04 | 2.32 | 3.71 | 2.85 | |
| Llama-3-70BBackbone=Llama-3-70B2026.04 | 0.71 | 7.14 | 3.18 | |
| BinderBackbone=GPT-3.52026.04 | 0 | 0 | 0 |