sui-1: Grounded and Verifiable Long-Form Summarization

TL;DR

sui-1 is a 24B parameter model that generates verifiable, citation-grounded summaries across multiple languages, significantly outperforming larger models by combining synthetic data and multi-stage verification.

Contribution

introduces sui-1, a model that produces traceable summaries with inline citations, leveraging a novel synthetic data pipeline and verification process for improved faithfulness.

Findings

sui-1 outperforms larger open-weight models in citation-grounded summarization

task-specific training surpasses scale in model performance

model and demo are publicly available

Abstract

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.