Web-based tool-use on Mind2Web
39.84Task PerformanceLAR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LARBackbone=Qwen3-8B2026.05 | 39.84 | -2.9 | |
| VanillaBackbone=Qwen3-8B2026.05 | 36.73 | — | |
| ConciseHintBackbone=Qwen3-8B2026.05 | 35.33 | 11.9 | |
| COTBackbone=Qwen3-8B2026.05 | 34.15 | — | |
| TokenSkipBackbone=Qwen3-8B2026.05 | 31.13 | -16.6 | |
| ACONBackbone=Qwen3-8B2026.05 | 30.77 | -17.7 | |
| LARBackbone=Llama-3.1 8B-Instruct2026.05 | 28.3 | -20.8 | |
| VanillaBackbone=Llama-3.1 8B-Instruct2026.05 | 24.4 | — | |
| ConciseHintBackbone=Llama-3.1 8B-Instruct2026.05 | 17.33 | -13 | |
| ACONBackbone=Llama-3.1 8B-Instruct2026.05 | 15.63 | -16.7 | |
| TokenSkipBackbone=Llama-3.1 8B-Instruct2026.05 | 14.27 | -1.9 | |
| COTBackbone=Llama-3.1 8B-Instruct2026.05 | 13.85 | — |