In-context retrieval on Real-world retrieval tasks 2K tokens (test)
41.96SWDE ScoreGated DeltaNet-2
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Gated DeltaNet-2Architecture Category=Attention or hybrid models, Input length=2K tokens2026.05 | 41.96 | 44.7 | 54.68 | 62.38 | 26.31 | 23.67 | 42.28 | |
| KDAArchitecture Category=Attention or hybrid models, Input length=2K tokens2026.05 | 39.83 | 40.1 | 53.59 | 59.89 | 25.27 | 22.18 | 40.14 | |
| Mamba-3 (SISO)Architecture Category=Attention or hybrid models, Input length=2K tokens, Configuration=SISO2026.05 | 35.3 | 46.42 | 54.95 | 59.54 | 25.91 | 23.96 | 41.01 | |
| Mamba-2Architecture Category=Attention or hybrid models, Input length=2K tokens2026.05 | 34.67 | 40.74 | 52.31 | 60.13 | 25.91 | 24.68 | 39.74 | |
| Gated DeltaNetArchitecture Category=Attention or hybrid models, Input length=2K tokens2026.05 | 33.18 | 42.28 | 50.86 | 60.6 | 25.78 | 21.95 | 39.11 | |
| Mamba-3 (MIMO)Architecture Category=Attention or hybrid models, Input length=2K tokens, Configuration=MIMO2026.05 | 32.33 | 44.7 | 55.31 | 59 | 26.26 | 23.08 | 40.11 | |
| TransformerArchitecture Category=Attention or hybrid models, Input length=2K tokens2026.05 | 32.21 | 38.67 | 54.78 | 58.09 | 22.49 | 22.18 | 38.07 | |
| Gated DeltaNet-2Architecture Category=Recurrent models, Input length=2K tokens2026.05 | 23.65 | 36.75 | 19.98 | 61.37 | 19.64 | 17.87 | 29.88 | |
| KDAArchitecture Category=Recurrent models, Input length=2K tokens2026.05 | 22.49 | 35.1 | 14.9 | 58.12 | 19.58 | 21.8 | 28.67 | |
| Gated DeltaNetArchitecture Category=Recurrent models, Input length=2K tokens2026.05 | 17.9 | 32.67 | 18.52 | 59.6 | 20.16 | 19.69 | 28.09 | |
| Mamba-3 (SISO)Architecture Category=Recurrent models, Input length=2K tokens, Configuration=SISO2026.05 | 17.62 | 35.07 | 11.08 | 58.89 | 18.18 | 21.32 | 27.03 | |
| Mamba-2Architecture Category=Recurrent models, Input length=2K tokens2026.05 | 17.24 | 32.38 | 14.53 | 58.35 | 18.91 | 19.6 | 26.84 | |
| Mamba-3 (MIMO)Architecture Category=Recurrent models, Input length=2K tokens, Configuration=MIMO2026.05 | 16.68 | 36.65 | 17.44 | 59.06 | 19.16 | 21.08 | 28.35 |