Agentic Search on Search Unseen
52.3PopQA AccuracyRESKILL
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| RESKILLModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 52.3 | 68.7 | 47.6 | 22.6 | 50.4 | 48 | 49.8 | |
| GRPOModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 51.8 | 63.7 | 40.7 | 16.2 | 43.5 | 43.1 | 45.3 | |
| EvolveRModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 51.2 | 65.7 | 43.9 | 16.7 | 45.6 | 44.5 | 46.6 | |
| MemRL (+RL)Model Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 49.8 | 65.7 | 41.2 | 19 | 42.1 | 43.8 | 46.2 | |
| RESKILLModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 49.7 | 66.9 | 45.7 | 18.3 | 47.5 | 45.4 | 47.6 | |
| SkillRLModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 48.5 | 65 | 39.7 | 15.3 | 44.8 | 42.4 | 45.1 | |
| MemRL (+RL)Model Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 48.2 | 65.3 | 35.8 | 15.9 | 44.3 | 41.9 | 44.2 | |
| INSPOModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 48.2 | 64.7 | 39.4 | 16.9 | 42.1 | 42.3 | 45 | |
| SkillRLModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 47.2 | 64.7 | 47.4 | 20.8 | 48.3 | 45.3 | 47.5 | |
| EvolveRModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 46.6 | 65.7 | 43.4 | 16.7 | 43.2 | 43.1 | 45 | |
| INSPOModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 46.5 | 63 | 44 | 15.8 | 42.4 | 42.3 | 45.1 | |
| Sonnet 4.5Model Scale=Sonnet 4.5, Strategy=Proprietary2026.06 | 43.8 | 67.8 | 31.1 | 17.6 | 57.1 | 41.7 | 38.6 | |
| GRPOModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 42.9 | 66.7 | 40.9 | 18.3 | 40.3 | 42 | 44.6 | |
| Best†Model Scale=Qwen3-8B, Strategy=w/ Skill2026.06 | 42.6 | 57 | 30.6 | 11.2 | 40.8 | 35.5 | 35.2 | |
| Best†Model Scale=Qwen3-4B-Instruct, Strategy=w/ Skill2026.06 | 42.1 | 55.1 | 30.9 | 10.8 | 42.1 | 34.9 | 34.5 | |
| Best‡Model Scale=Qwen3-4B-Instruct, Strategy=w/ Evolve2026.06 | 39.4 | 57.3 | 34.9 | 11.1 | 41.3 | 35.4 | 34.5 | |
| Best‡Model Scale=Qwen3-8B, Strategy=w/ Evolve2026.06 | 37.5 | 57.2 | 27.5 | 11.9 | 40.3 | 34.1 | 33.8 | |
| ReActModel Scale=Qwen3-8B, Strategy=Baseline2026.06 | 37.2 | 52.1 | 30.8 | 9.8 | 35.2 | 32.8 | 32.2 | |
| ReActModel Scale=Qwen3-4B-Instruct, Strategy=Baseline2026.06 | 29.4 | 40.7 | 30.7 | 9.3 | 30.4 | 27.8 | 27.2 |