Agentic Tool-Use on BFCL
94.1BFCL Average ScoreGEAR
Evaluation results
| Method | Links | |
|---|---|---|
| GEARBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 94.1 | |
| GEARBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 93.8 | |
| ARPOBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 92.6 | |
| MT-GRPOBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 92.4 | |
| ARPOBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 92.3 | |
| MT-GRPOBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 92.3 | |
| BaseBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 92.1 | |
| GRPOBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 92 | |
| GRPOBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 91.9 | |
| GEARBackbone=Qwen3-4B, Evaluation Context=Multi-Domain Generalization2026.05 | 91.9 | |
| GRPOBackbone=Qwen3-4B, Evaluation Context=Multi-Domain Generalization2026.05 | 91.1 | |
| BaseBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 90.7 | |
| OPSD+RLBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 88.5 | |
| OPSD+RLBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 87.1 | |
| OPSDBackbone=Qwen3-8B, Evaluation Context=Standard2026.05 | 76.4 | |
| OPSDBackbone=Qwen3-4B, Evaluation Context=Standard2026.05 | 72 |