Agent tool-use on ClawEval 114 T-family tasks
37.7p@1SFT ISETrace
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SFT ISETracetraining=Supervised Fine-Tuning (SFT), base_model=Qwen3-8B2026.06 | 37.7 | 53.3 | 95.9 | |
| Qwen3-32Bvariant=base, prompting=0-shot2026.06 | 30.7 | 44.6 | 94.7 | |
| GPT-4oprompting=0-shot2026.06 | 25.4 | 44.2 | 96.5 | |
| Qwen3-8Bvariant=base, prompting=0-shot2026.06 | 19.3 | 36.7 | 92.5 |