Long-context retrieval on Lost-in-the-Middle 30-passage contexts
62.57Average Exact MatchPRISM-∆
Evaluation results
| Method | Links | |
|---|---|---|
| PRISM-∆Model=QWEN3-14B2026.03 | 62.57 | |
| PRISM-∆VModel=QWEN3-14B2026.03 | 62 | |
| SEKAModel=QWEN3-14B2026.03 | 60.86 | |
| VanillaModel=QWEN3-14B2026.03 | 57.13 | |
| SEKAModel=GEMMA3-12B2026.03 | 55.57 | |
| PRISM-∆VModel=GEMMA3-12B2026.03 | 55.57 | |
| PRISM-∆Model=GEMMA3-12B2026.03 | 54.43 | |
| PRISM-∆Model=QWEN3-8B2026.03 | 52.71 | |
| PRISM-∆VModel=QWEN3-8B2026.03 | 51.86 | |
| VanillaModel=GEMMA3-12B2026.03 | 51.64 | |
| PRISM-∆Model=QWEN3-4B2026.03 | 51.43 | |
| PRISM-∆VModel=QWEN3-4B2026.03 | 50.57 | |
| SEKAModel=QWEN3-8B2026.03 | 50.29 | |
| PRISM-∆VModel=GEMMA3-4B2026.03 | 50.14 | |
| PRISM-∆Model=GEMMA3-4B2026.03 | 49.29 | |
| SEKAModel=QWEN3-4B2026.03 | 49.14 | |
| SEKAModel=GEMMA3-4B2026.03 | 47.71 | |
| VanillaModel=QWEN3-8B2026.03 | 46.69 | |
| VanillaModel=QWEN3-4B2026.03 | 45.47 | |
| VanillaModel=GEMMA3-4B2026.03 | 44.36 |