Agent Skill Performance on SpreadsheetBench
69.83Vrf ScoreTrace2Skill (Human-Written + Combined)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Trace2Skill (Human-Written + Combined)Skill User=122B, Deepening setting=same-model Deepening2026.03 | 69.83 | 47.17 | 29.53 | |
| ReasoningBankSkill User=122B, Deepening setting=same-model Deepening2026.03 | 56 | 40.1 | 21.3 | |
| Trace2Skill (Human-Written + Combined)Skill User=35B, Deepening setting=same-model Deepening2026.03 | 29.67 | 18.8 | 5.73 | |
| ReasoningBankSkill User=35B, Deepening setting=same-model Deepening2026.03 | 20.5 | 17.3 | 4.97 |