Function Calling on BFCL Live v3
77.9Overall AccuracyQwen3-4B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3-4BSpeedup ratio=1×2026.02 | 77.9 | 94.5 | |
| Qwen2.5-14BSpeedup ratio=1×2026.02 | 76.6 | 95.9 | |
| ST-Qwen3-4BSpeedup ratio=3.9×2026.02 | 76.4 | 96.3 | |
| ST-Qwen2.5-14BSpeedup ratio=4.4×2026.02 | 75.9 | 97.1 | |
| ST-Qwen2.5-7BSpeedup ratio=4.5×2026.02 | 75.8 | 96.7 | |
| Qwen2.5-7BSpeedup ratio=1×2026.02 | 75.4 | 95.2 | |
| Qwen2.5-3BSpeedup ratio=1×2026.02 | 73 | 94.3 | |
| ST-Qwen2.5-3BSpeedup ratio=3.8×2026.02 | 68 | 95.4 | |
| Qwen2.5-1.5BSpeedup ratio=1×2026.02 | 67.4 | 90.4 | |
| ST-Qwen2.5-1.5BSpeedup ratio=3.5×2026.02 | 63.6 | 94.3 | |
| ST-Qwen2.5-0.5BSpeedup ratio=3.1×2026.02 | 57.2 | 91.6 | |
| Qwen2.5-0.5BSpeedup ratio=1×2026.02 | 49.9 | 86.4 |