Retrieval on 100 questions cross-model comparison
41.2F1 ScoreRLM-on-KG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RLM-on-KGController=Claude Haiku 4.52026.04 | 41.2 | — | |
| RLM-on-KGController=Gemini 2.5 Flash Lite2026.04 | 38 | 68 | |
| Heuristic RLMController=Claude Haiku 4.52026.04 | 37.6 | — | |
| Heuristic RLMController=Gemini 2.5 Flash Lite2026.04 | 37.1 | 68 | |
| Heuristic RLMController=Gemma 4 E2B (local)2026.04 | 35 | 237 | |
| RLM-on-KGController=Gemma 4 E2B (local)2026.04 | 34.3 | 237 |