Tool Use on Unified evaluation benchmark D Single-Hop (test)
96.1SP ScoreQwen3-8B (Upper Bound)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3-8B (Upper Bound)Setting=GT Setting2026.04 | 96.1 | 96.1 | 28.7 | 52.3 | |
| UniToolCallSetting=Hybrid-20 Setting2026.04 | 92.9 | 92.9 | 27.1 | 48.6 | |
| Qwen3-32BSetting=Hybrid-20 Setting2026.04 | 72.6 | 72.6 | 23.7 | 43.8 | |
| Gemini 3 Flash PreviewSetting=Hybrid-20 Setting2026.04 | 68.3 | 68.3 | 25.2 | 46.3 | |
| Qwen3-8B (Vanilla)Setting=Hybrid-20 Setting2026.04 | 66.9 | 66.9 | 19.8 | 19.8 | |
| Claude 4.6 SonnetSetting=Hybrid-20 Setting2026.04 | 58.8 | 58.8 | 24.9 | 43.1 | |
| Kimi-K2-InstructSetting=Hybrid-20 Setting2026.04 | 55.9 | 55.9 | 21.3 | 37.1 | |
| GPT-5.2 InstantSetting=Hybrid-20 Setting2026.04 | 50.9 | 50.9 | 23.2 | 39.2 | |
| DeepSeek-V3.2Setting=Hybrid-20 Setting2026.04 | 46.9 | 46.9 | 18.8 | 32.8 |