Metraipolis › archive › dc_0d2ff1401a51
YANchor-4B: Effective Long-Horizon Reasoning in O(N) Time with O(1) Memory — arXiv:2610.10118
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.10118
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T11:26:39.342Z
- SHA-256 of the archived text
e97f05acdd31e4e6562db9cbc92208c8c5621157b5b9103e4e410a79e3234b88- Stable URL of this page
https://metraipolis.com/documento/dc_0d2ff1401a51
How to cite
YANchor-4B: Effective Long-Horizon Reasoning in O(N) Time with O(1) Memory — arXiv:2610.10118 — arXiv (en), https://arxiv.org/abs/2610.10118 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_0d2ff1401a51 (Metraipolis archive dc_0d2ff1401a51, sha256 e97f05acdd31e4e6562db9cbc92208c8c5621157b5b9103e4e410a79e3234b88)
BibTeX
@misc{dc0d2ff1401a51,
title = {YANchor-4B: Effective Long-Horizon Reasoning in O(N) Time with O(1) Memory — arXiv:2610.10118},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.10118},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_0d2ff1401a51; sha256 e97f05acdd31e4e6562db9cbc92208c8c5621157b5b9103e4e410a79e3234b88},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.10118. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Long-horizon reasoning demands access to earlier information at a manageable generation cost. Full-history attention incurs growing storage and computation, while recurrent compression can lose precise details. Therefore, we present YANchor-4B, a general-purpose recurrent model that preserves crucial memory as ANchors for retrieval during subsequent reasoning. Beyond $O(N)$-time generation and $O(1)$ memory, YANchor enables effective long-horizon reasoning through its multidimensional memory mechanism. For example, on challenging math problems, it achieves 82.93% mean pass@1 on AIME 2024--2026 and 63.64% on HMMT, substantially outperforming linear-time, constant-state counterparts, including larger models. It also delivers several-fold higher batched long-generation throughput than Transformer and hybrid baselines on H100. Furthermore, evaluations across dozens of benchmarks demonstrate YANchor's superiority in general-purpose capabilities. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2610.10118 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-08T10:24:06.391Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).