Long-context retrieval on RULER Multi-Key (test)
99.5Accuracy (4K Context)Qwen3-1.7B
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen3-1.7BType=FA2026.06 | 99.5 | 99.4 | 97.8 | 94.1 | 42.2 | 0 | 0 | |
| Gemma-3n-E2BType=Hybrid2026.06 | 98.7 | 98.5 | 89.7 | 76.3 | 11.7 | 0 | 0 | |
| HydraHeadType=Hybrid2026.06 | 98.5 | 96.4 | 94.3 | 92.5 | 82 | 68.8 | 52.7 | |
| MiniCPM5-1B-BaseType=FA2026.06 | 98.1 | 96.6 | 89.9 | 87.2 | 76.2 | 54.4 | 16.6 | |
| Qwen3-1.7B-yarnType=FA2026.06 | 97.4 | 91.8 | 54.3 | 46.1 | 47.8 | 41.5 | 14.2 | |
| Jet-Nemotron-2BType=Hybrid2026.06 | 95.7 | 89.9 | 69 | 42 | 8.8 | 1.9 | 0 | |
| Hymba-1.5BType=Hybrid2026.06 | 44.1 | 33.4 | 12.5 | 0 | 0 | 0 | 0 | |
| HypeNet-2BType=Hybrid2026.06 | 30.5 | 28.9 | 25 | 13.9 | 27.4 | 16 | 16.9 |