sui-1: Grounded and Verifiable Long-Form Summarization

2026-01-13

AI Summary

This work addresses the lack of verifiability in summaries generated by large language models, which poses significant trustworthiness risks in compliance-sensitive domains such as government and legal applications. To mitigate this issue, the authors propose a 24-billion-parameter verifiable summarization model that, for the first time, enables fine-grained citation tracing in multilingual long-document summarization—each summary span is explicitly linked to its source sentence in the original text. The model is trained using chain-of-thought prompting, a multi-stage automatic verification mechanism, and a synthetic data pipeline, drawing from diverse multilingual sources including parliamentary proceedings, web pages, and Wikipedia across five languages. Experimental results demonstrate that the proposed approach substantially outperforms all open-source baselines, including models with three times its parameter count. Model weights and an interactive demo are publicly released.

Technology Category

• Natural Language Processing: Summarization
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Web Mining and Content Analysis: Large pretrained models with web data

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.

Problem

• grounded summarization
• verifiable summaries
• citation
• faithfulness
• compliance-sensitive domains

Innovation

• citation-grounded summarization
• inline citations
• synthetic data pipeline
• chain-of-thought prompting
• verifiable abstractive summarization