Spreadsheet Reasoning on SpreadsheetBench Verified
71.11AccuracySkillGrad
Evaluation Results
| Method | Links | |
|---|---|---|
| SkillGradBackbone=GPT-5.4, Initialization=LLM-gen., Training Type=Training-based skill improvement2026.05 | 71.11 | |
| SkillGradBackbone=GPT-5.4, Initialization=Third-party, Training Type=Training-based skill improvement2026.05 | 69.44 | |
| EvoSkillBackbone=GPT-5.4, Initialization=LLM-gen., Training Type=Training-based skill improvement2026.05 | 68.06 | |
| Trace2SkillBackbone=GPT-5.4, Initialization=LLM-gen., Training Type=Training-based skill improvement2026.05 | 65.28 | |
| Trace2SkillBackbone=GPT-5.4, Initialization=Third-party, Training Type=Training-based skill improvement2026.05 | 63.89 | |
| EvoSkillBackbone=GPT-5.4, Initialization=Third-party, Training Type=Training-based skill improvement2026.05 | 63.61 | |
| No SkillBackbone=GPT-5.4, Initialization=None, Training Type=Training-free2026.05 | 62.5 | |
| Base xlsx SkillBackbone=GPT-5.4, Initialization=Third-party, Training Type=Training-free2026.05 | 60 | |
| Base xlsx SkillBackbone=GPT-5.4, Initialization=LLM-gen., Training Type=Training-free2026.05 | 55.83 | |
| SkillGradBackbone=GPT-4.1, Initialization=LLM-gen., Training Type=Training-based skill improvement2026.05 | 54.17 | |
| SkillGradBackbone=GPT-4.1, Initialization=Third-party, Training Type=Training-based skill improvement2026.05 | 45.83 | |
| No SkillBackbone=GPT-4.1, Initialization=None, Training Type=Training-free2026.05 | 44.17 | |
| Trace2SkillBackbone=GPT-4.1, Initialization=Third-party, Training Type=Training-based skill improvement2026.05 | 38.89 | |
| Trace2SkillBackbone=GPT-4.1, Initialization=LLM-gen., Training Type=Training-based skill improvement2026.05 | 37.22 | |
| EvoSkillBackbone=GPT-4.1, Initialization=LLM-gen., Training Type=Training-based skill improvement2026.05 | 37.22 | |
| EvoSkillBackbone=GPT-4.1, Initialization=Third-party, Training Type=Training-based skill improvement2026.05 | 36.94 | |
| Base xlsx SkillBackbone=GPT-4.1, Initialization=LLM-gen., Training Type=Training-free2026.05 | 36.67 | |
| Base xlsx SkillBackbone=GPT-4.1, Initialization=Third-party, Training Type=Training-free2026.05 | 33.33 |