Composing Auto-Generated Skills on IW
30Skill-Step PercentageLlama-3.1-70B
Evaluation results
| Method | Links | |
|---|---|---|
| Llama-3.1-70BEvaluation Protocol=zero-shot format-constrained prompting2026.06 | 30 | |
| Claude Haiku 4.5Evaluation Protocol=zero-shot format-constrained prompting2026.06 | 27.8 | |
| Claude Sonnet 4.5Evaluation Protocol=zero-shot format-constrained prompting2026.06 | 25.8 | |
| GPT-5Evaluation Protocol=zero-shot format-constrained prompting2026.06 | 24.5 | |
| Qwen3-8BEvaluation Protocol=GRPO from base2026.06 | 20.5 | |
| Qwen3-8BEvaluation Protocol=zero-shot2026.06 | 18.5 | |
| Gemma-4-31BEvaluation Protocol=zero-shot format-constrained prompting2026.06 | 16.5 | |
| OLMo-3-7BEvaluation Protocol=zero-shot format-constrained prompting2026.06 | 14.3 | |
| DeepSeek-R1-Distill-32BEvaluation Protocol=zero-shot format-constrained prompting2026.06 | 14.3 |