Multi-turn tool-use interaction on VitaBench
59Delivery ScoreU-Fold
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| U-FoldModel=Claude-4.5-Sonnet2026.01 | 59 | 64 | 48.8 | 38 | — | |
| U-FoldModel=DeepSeek-V3.2-Exp2026.01 | 56.8 | 58 | 21 | 21 | — | |
| ReActModel=Claude-4.5-Sonnet2026.01 | 55.8 | 56 | 45.5 | 35.5 | 2.8 | |
| ReSumModel=Claude-4.5-Sonnet2026.01 | 53.8 | 55 | 34 | 35.3 | 7.3 | |
| U-FoldModel=GPT-4.12026.01 | 51.3 | 58.5 | 33 | 22.3 | — | |
| IterResearchModel=Claude-4.5-Sonnet2026.01 | 50.8 | 59.8 | 30 | 27 | 9.5 | |
| ReActModel=DeepSeek-V3.2-Exp2026.01 | 48.5 | 56.3 | 17.5 | 20 | 3.4 | |
| ReActModel=GPT-4.12026.01 | 45.5 | 55.8 | 29 | 17.3 | 2.5 | |
| ReSumModel=GPT-4.12026.01 | 40 | 46.5 | 25.8 | 21.3 | 12.7 | |
| U-FoldModel=Qwen3-Thinking-30B-A3B2026.01 | 35 | 44 | 13.8 | 19.5 | — | |
| ReActModel=Qwen3-Thinking-30B-A3B2026.01 | 33.3 | 43.5 | 11.7 | 15 | 1.5 | |
| ReSumModel=Qwen3-Thinking-30B-A3B2026.01 | 32 | 41.3 | 9.3 | 9.3 | 6.2 | |
| ReSumModel=DeepSeek-V3.2-Exp2026.01 | 31.5 | 20.3 | 5 | 8.3 | 15.8 | |
| IterResearchModel=DeepSeek-V3.2-Exp2026.01 | 29.5 | 28.8 | 15 | 11.8 | 27 | |
| IterResearchModel=GPT-4.12026.01 | 26 | 38.5 | 16 | 8.8 | 13.9 | |
| IterResearchModel=Qwen3-Thinking-30B-A3B2026.01 | 24.5 | 41.8 | 12 | 10.8 | 11.3 | |
| IterResearchModel=Qwen3-4B2026.01 | 13 | 7 | 0 | 0 | 3.7 | |
| U-FoldModel=Qwen3-4B2026.01 | 10 | 8 | 0 | 2 | — | |
| ReActModel=Qwen3-4B2026.01 | 9.3 | 5 | 0 | 0 | 1.1 | |
| ReSumModel=Qwen3-4B2026.01 | 5 | 2 | 0 | 0 | 3.7 |