Function Calling on BFCL v4 (Base/Error Metrics)
72.4Base AccuracyTRACE
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TRACEModel=Qwen3-14B2026.06 | 72.4 | 34.8 | 44.6 | 40.2 | 48 | |
| TreePOModel=Qwen3-14B2026.06 | 70.8 | 33 | 43.2 | 39.4 | 46.6 | |
| GRPOModel=Qwen3-14B2026.06 | 70.6 | 32.6 | 40.9 | 36.4 | 46.1 | |
| PCLModel=Qwen3-14B2026.06 | 70 | 31.8 | 41.4 | 40.4 | 45.9 | |
| TRACEModel=Qwen3-8B2026.06 | 61.2 | 34.4 | 48.8 | 40.4 | 46.2 | |
| PCLModel=Qwen3-8B2026.06 | 59 | 32.1 | 46.8 | 39.3 | 44.3 | |
| TreePOModel=Qwen3-8B2026.06 | 58.9 | 32 | 46.5 | 39.4 | 44.2 | |
| GRPOModel=Qwen3-8B2026.06 | 58.5 | 31 | 45.8 | 38.7 | 43.5 | |
| ReActModel=Qwen3-14B2026.06 | 51.1 | 23.6 | 35.6 | 22.8 | 33.6 | |
| ReActModel=Qwen3-8B2026.06 | 36.7 | 21.5 | 31.1 | 22.3 | 28 |