Goal-conditioned tool use on Aggregate Downstream Tasks (test)
99.5Success RateGold-goal CMTF
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Gold-goal CMTFFiltering Method=Gold-goal CMTF2026.06 | 99.5 | 0 | 1 | 0.332 | 0 | 16.7 | 0 | 689 | |
| GIST-CMTFFiltering Method=GIST-CMTF2026.06 | 97 | 2.5 | 1 | 1.051 | 0 | 66 | 5.6 | 1,186 | |
| Semantic-goal CMTFFiltering Method=Semantic-goal CMTF2026.06 | 82.9 | 16.7 | 1 | 0.533 | 0 | 28.3 | 0 | 700 | |
| Top-goal CMTFFiltering Method=Top-goal CMTF2026.06 | 80.1 | 19.4 | 1 | 0.642 | 0 | 33.1 | 0 | 1,077 | |
| State-awareFiltering Method=State-aware2026.06 | 56.4 | 0 | 8.123 | 0.844 | 0 | 46.8 | 1.7 | 2,084 | |
| All toolsFiltering Method=All tools2026.06 | 53.5 | 0 | 32 | 0.788 | 3.539 | 33.7 | 0 | 4,152 |