Tool Use on Unified evaluation benchmark D Single-Turn (test)
93SPUniToolCall
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| UniToolCallSetting=Hybrid-20 Setting2026.04 | 93 | 93.8 | 31.6 | 52.4 | |
| Qwen3-8B (Upper Bound)Setting=GT Setting2026.04 | 92.6 | 95.5 | 32.1 | 54.9 | |
| Qwen3-32BSetting=Hybrid-20 Setting2026.04 | 72.7 | 74.3 | 27.9 | 47.4 | |
| Gemini 3 Flash PreviewSetting=Hybrid-20 Setting2026.04 | 70.3 | 70.9 | 30.3 | 50.5 | |
| Qwen3-8B (Vanilla)Setting=Hybrid-20 Setting2026.04 | 63.3 | 66.5 | 22.1 | 22.1 | |
| Claude 4.6 SonnetSetting=Hybrid-20 Setting2026.04 | 62.1 | 62.7 | 30.3 | 47.9 | |
| Kimi-K2-InstructSetting=Hybrid-20 Setting2026.04 | 58.7 | 59.7 | 26.5 | 42 | |
| GPT-5.2 InstantSetting=Hybrid-20 Setting2026.04 | 50.5 | 52.5 | 26.3 | 41.5 | |
| DeepSeek-V3.2Setting=Hybrid-20 Setting2026.04 | 47 | 49.6 | 22.9 | 36.3 |