sui-1: Grounded and Verifiable Long-Form Summarization

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Droste, Benedikt, Harries, Jan Philipp, Idahl, Maximilian, Plüster, Björn
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914251624415232
author Droste, Benedikt
Harries, Jan Philipp
Idahl, Maximilian
Plüster, Björn
author_facet Droste, Benedikt
Harries, Jan Philipp
Idahl, Maximilian
Plüster, Björn
contents Large language models frequently generate plausible but unfaithful summaries that users cannot verify against source text, a critical limitation in compliance-sensitive domains such as government and legal analysis. We present sui-1, a 24B parameter model that produces abstractive summaries with inline citations, enabling users to trace each claim to its source sentence. Our synthetic data pipeline combines chain-of-thought prompting with multi-stage verification, generating over 22,000 high-quality training examples across five languages from diverse sources including parliamentary documents, web text, and Wikipedia. Evaluation shows sui-1 significantly outperforms all tested open-weight baselines, including models with 3x more parameters. These results demonstrate that task-specific training substantially outperforms scale alone for citation-grounded summarization. Model weights and an interactive demo are publicly available.
format Preprint
id arxiv_https___arxiv_org_abs_2601_08472
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle sui-1: Grounded and Verifiable Long-Form Summarization
Droste, Benedikt
Harries, Jan Philipp
Idahl, Maximilian
Plüster, Björn
Computation and Language
Artificial Intelligence
Large language models frequently generate plausible but unfaithful summaries that users cannot verify against source text, a critical limitation in compliance-sensitive domains such as government and legal analysis. We present sui-1, a 24B parameter model that produces abstractive summaries with inline citations, enabling users to trace each claim to its source sentence. Our synthetic data pipeline combines chain-of-thought prompting with multi-stage verification, generating over 22,000 high-quality training examples across five languages from diverse sources including parliamentary documents, web text, and Wikipedia. Evaluation shows sui-1 significantly outperforms all tested open-weight baselines, including models with 3x more parameters. These results demonstrate that task-specific training substantially outperforms scale alone for citation-grounded summarization. Model weights and an interactive demo are publicly available.
title sui-1: Grounded and Verifiable Long-Form Summarization
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2601.08472