Function Calling on BFCL Multi-turn V4
60Base Success RateAgenticQwen-30B-A3B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AgenticQwen-30B-A3BReasoning Method=non-thinking, using tools2026.04 | 60 | 52 | 29 | 55.5 | |
| Qwen3-235B-A22B-InstructReasoning Method=non-thinking, using tools2026.04 | 58.5 | 47.5 | 35 | 54 | |
| AgenticQwen-8BReasoning Method=non-thinking, using tools2026.04 | 56 | 47.5 | 33.5 | 40.5 | |
| Qwen3-32BReasoning Method=non-thinking, using tools2026.04 | 50.5 | 43 | 30.5 | 33 | |
| Qwen3-30B-A3B-InstructReasoning Method=non-thinking, using tools2026.04 | 47 | 14 | 28 | 45.5 | |
| Qwen3-8BReasoning Method=non-thinking, using tools2026.04 | 35.5 | 35 | 20.5 | 21.5 |