| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| Real-world data | StaticHybrid | SQuAD Accuracy50.1 | 30 | 2mo ago | |
| In-context retrieval (lm-evaluation-harness) zero-shot | KDA | FDA Accuracy23.23 | 23 | 3mo ago | |
| DROP | BERT-Judge | Accuracy88.6 | 16 | 3mo ago | |
| RULER | Mamba-2 | S-NIAH-1 (1K)100 | 13 | 29d ago | |
| Real-world retrieval tasks 2K tokens (test) | Gated DeltaNet-2 | SWDE Score41.96 | 13 | 2mo ago | |
| SWDE | Hybrid Gated DeltaNet + M2RNN-1 | Accuracy62.5 | 13 | 4mo ago | |
| FDA | Hybrid M2RNN | Accuracy74.5 | 13 | 4mo ago | |
| TriviaQA | Hybrid Mamba-2 + M2RNN-3 | Accuracy56.7 | 13 | 4mo ago | |
| SQuAD | Hybrid M2RNN | Accuracy41.3 | 13 | 4mo ago | |
| RULER 16K | Full-Attn HoPE (extra MLP) | S-N Score100 | 10 | 18d ago | |
| RULER 8K | Full-Attn RoPE | S-N100 | 10 | 18d ago | |
| RULER (test) | S1 Score100 | 8 | 3mo ago | ||
| RULER 32K | HiLS-Attn-HoPE | S-N Score100 | 7 | 18d ago | |
| MAD (test) | Hymba w/ MATCH | Fuzzy ICR80.3 | 6 | 25d ago | |
| RULER | MQ Score99.55 | 4 | 3mo ago | ||
| RULER 512K | HiLS-Attn-HoPE | S-N Score99 | 3 | 18d ago | |
| RULER 128K | HiLS-Attn-HoPE | S-N Score99 | 3 | 18d ago | |
| RULER 4M | HiLS-Attn-HoPE | S-N Score96 | 1 | 18d ago | |
| RULER 2M | HiLS-Attn-HoPE | S-N97 | 1 | 18d ago | |
| RULER 1M | HiLS-Attn-HoPE | S-N Score100 | 1 | 18d ago |