End-to-end Search Agent on BrowseComp-Plus
51.45AccuracyMPE-Rand
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MPE-RandBackbone=Qwen3-Embedding-0.6B, Training Range=[32, 1024], Search Agent=Gemini 3 Flash Preview, Evaluator=GPT-4o-mini, Inference Chunk Size=642026.06 | 51.45 | 57.7 | 15.78 | 39.74 | |
| Single-vectorBackbone=Qwen3-Embedding-0.6B, Search Agent=Gemini 3 Flash Preview, Evaluator=GPT-4o-mini, Inference Chunk Size=642026.06 | 42.29 | 48 | 17.4 | 46.99 | |
| BM25Search Agent=Gemini 3 Flash Preview, Evaluator=GPT-4o-mini2026.06 | 34.94 | 37.78 | 18.85 | 51.17 |