Composing auto-generated skills on BrowseComp+
61.4Skill Step PercentageOLMo-3-7B
Evaluation results
| Method | Links | |
|---|---|---|
| OLMo-3-7BEvaluation Protocol=zero-shot format-constrained prompting2026.06 | 61.4 | |
| GPT-5Evaluation Protocol=zero-shot format-constrained prompting2026.06 | 59.5 | |
| Llama-3.1-70BEvaluation Protocol=zero-shot format-constrained prompting2026.06 | 51.9 | |
| Claude Haiku 4.5Evaluation Protocol=zero-shot format-constrained prompting2026.06 | 51 | |
| Claude Sonnet 4.5Evaluation Protocol=zero-shot format-constrained prompting2026.06 | 47.7 | |
| Qwen3-8BEvaluation Protocol=zero-shot2026.06 | 43.5 | |
| Qwen3-8BEvaluation Protocol=GRPO from base2026.06 | 43.3 | |
| Gemma-4-31BEvaluation Protocol=zero-shot format-constrained prompting2026.06 | 11.2 | |
| DeepSeek-R1-Distill-32BEvaluation Protocol=zero-shot format-constrained prompting2026.06 | 0.05 |