cnn-/-daily-mail.

BRIO: Bringing Order to Abstractive Summarization. ACL 2022. BART-large with contrastive learning. SOTA on CNN/DM at time of publication. Score from Table 1 of the paper.

From paper: SummaReranker: A Multi-Task Mixture-of-Experts Re-ranking Framework for Abstractive Summarization

GPT-3.5-Turbo prompted with TriSum structured rationale. Best ROUGE-1 in TriSum paper Table 2. Zero-shot with structured chain-of-thought prompting.

BRIDO: Bringing Democratic Order to Abstractive Summarization. arXiv Feb 2025. BART-based with democratic contrastive learning. Trades slight ROUGE drop vs BRIO for better factual consistency (3.82% G-Eval improvement). Table 2.

TriSum: Learning Summarization Ability from LLMs with Structured Rationale. NAACL 2024. TriSum-J = joint learning stage. Table 2.

From paper: Fourier Transformer: Fast Long Range Modeling by Removing Sequence Redundancy with FFT Operator

From paper: GLM: General Language Model Pretraining with Autoregressive Blank Infilling

From paper: Hierarchical Learning for Generation with Long Source Sequences

From paper: Extractive Summarization as Text Matching

From paper: Hie-BART: Document Summarization with Hierarchical BART

From paper: Extractive Summarization as Text Matching

From paper: Text Summarization with Pretrained Encoders

From paper: Big Bird: Transformers for Longer Sequences

From paper: Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer

SumHiS: Extractive Summarization Exploiting Hidden Structure. arXiv Jun 2024. With semantic filtering. Extractive approach; ROUGE-2 of 32.52 exceeds prior extractive SOTA by 10%. Table 1.

From paper: Fine-tune BERT for Extractive Summarization

From paper: Unified Language Model Pre-training for Natural Language Understanding and Generation

From paper: Mixture Content Selection for Diverse Sequence Generation

From paper: Neural Document Summarization by Jointly Learning to Score and Select Sentences

From paper: Bottom-Up Abstractive Summarization

Llama-2-70B-chat with 7-shot in-context learning on CNN/DailyMail. Best overall ICL result in arXiv:2507.05123 (Jul 2025). Outperforms zero-shot by substantial margin.

From paper: Time-aware Large Kernel Convolutions

From paper: Get To The Point: Summarization with Pointer-Generator Networks

From paper: Time-aware Large Kernel Convolutions

From paper: A Deep Reinforced Model for Abstractive Summarization

From paper: Pay Less Attention with Lightweight and Dynamic Convolutions

From paper: Pay Less Attention with Lightweight and Dynamic Convolutions

From paper: Synthesizer: Rethinking Self-Attention in Transformer Models

From paper: A Deep Reinforced Model for Abstractive Summarization

Zero-shot evaluation of Mistral-7B-Instruct-v0.1 on CNN/DailyMail. Best zero-shot 7B result in arXiv:2507.05123 (Jul 2025). Standard ROUGE scoring against reference highlights.

From paper: Coarse-to-Fine Attention Models for Document Summarization

From paper: Language Models are Unsupervised Multitask Learners

BRIO: Bringing Order to Abstractive Summarization. ACL 2022.

From paper: SummaReranker: A Multi-Task Mixture-of-Experts Re-ranking Framework for Abstractive Summarization

BRIDO: Bringing Democratic Order to Abstractive Summarization. arXiv Feb 2025. Table 2.

SumHiS: Extractive Summarization Exploiting Hidden Structure. arXiv Jun 2024. With semantic filtering. Table 1.

TriSum: Learning Summarization Ability from LLMs with Structured Rationale. NAACL 2024. TriSum-J = joint learning stage. Table 2.

From paper: Hierarchical Learning for Generation with Long Source Sequences

From paper: GLM: General Language Model Pretraining with Autoregressive Blank Infilling

From paper: Fourier Transformer: Fast Long Range Modeling by Removing Sequence Redundancy with FFT Operator

From paper: Hie-BART: Document Summarization with Hierarchical BART

From paper: Big Bird: Transformers for Longer Sequences

GPT-3.5-Turbo prompted with TriSum structured rationale. TriSum paper Table 2.

From paper: Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer

From paper: Extractive Summarization as Text Matching

From paper: Extractive Summarization as Text Matching

From paper: Unified Language Model Pre-training for Natural Language Understanding and Generation

From paper: Text Summarization with Pretrained Encoders

From paper: Fine-tune BERT for Extractive Summarization

From paper: Mixture Content Selection for Diverse Sequence Generation

From paper: Bottom-Up Abstractive Summarization

From paper: Neural Document Summarization by Jointly Learning to Score and Select Sentences

From paper: A Deep Reinforced Model for Abstractive Summarization

From paper: Time-aware Large Kernel Convolutions

From paper: Pay Less Attention with Lightweight and Dynamic Convolutions

From paper: Get To The Point: Summarization with Pointer-Generator Networks

From paper: Pay Less Attention with Lightweight and Dynamic Convolutions

From paper: Time-aware Large Kernel Convolutions

From paper: Synthesizer: Rethinking Self-Attention in Transformer Models

From paper: A Deep Reinforced Model for Abstractive Summarization

From paper: Coarse-to-Fine Attention Models for Document Summarization

Llama-2-70B-chat with 7-shot ICL on CNN/DailyMail. arXiv:2507.05123 (Jul 2025).

From paper: Language Models are Unsupervised Multitask Learners

Zero-shot evaluation of Mistral-7B-Instruct-v0.1 on CNN/DailyMail. arXiv:2507.05123 (Jul 2025).

From paper: Coarse-to-Fine Attention Models for Document Summarization

BRIO: Bringing Order to Abstractive Summarization. ACL 2022.

GPT-3.5-Turbo prompted with TriSum structured rationale. TriSum paper Table 2.

BRIDO: Bringing Democratic Order to Abstractive Summarization. arXiv Feb 2025. Table 2.

TriSum: Learning Summarization Ability from LLMs with Structured Rationale. NAACL 2024. TriSum-J = joint learning stage. Table 2.

From paper: SummaReranker: A Multi-Task Mixture-of-Experts Re-ranking Framework for Abstractive Summarization

From paper: Fourier Transformer: Fast Long Range Modeling by Removing Sequence Redundancy with FFT Operator

From paper: Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer

From paper: GLM: General Language Model Pretraining with Autoregressive Blank Infilling

From paper: Hie-BART: Document Summarization with Hierarchical BART

From paper: Hierarchical Learning for Generation with Long Source Sequences

From paper: Big Bird: Transformers for Longer Sequences

From paper: Extractive Summarization as Text Matching

From paper: Extractive Summarization as Text Matching

From paper: Unified Language Model Pre-training for Natural Language Understanding and Generation

From paper: Text Summarization with Pretrained Encoders

From paper: Fine-tune BERT for Extractive Summarization

From paper: Universal Evasion Attacks on Summarization Scoring

From paper: Neural Document Summarization by Jointly Learning to Score and Select Sentences

From paper: Time-aware Large Kernel Convolutions

From paper: Mixture Content Selection for Diverse Sequence Generation

From paper: Bottom-Up Abstractive Summarization

From paper: Time-aware Large Kernel Convolutions

From paper: Get To The Point: Summarization with Pointer-Generator Networks

Llama-2-70B-chat with 7-shot ICL on CNN/DailyMail. arXiv:2507.05123 (Jul 2025).

Zero-shot evaluation of Mistral-7B-Instruct-v0.1 on CNN/DailyMail. arXiv:2507.05123 (Jul 2025).

From paper: Pay Less Attention with Lightweight and Dynamic Convolutions

From paper: Synthesizer: Rethinking Self-Attention in Transformer Models

From paper: Pay Less Attention with Lightweight and Dynamic Convolutions

From paper: A Deep Reinforced Model for Abstractive Summarization

From paper: Coarse-to-Fine Attention Models for Document Summarization

From paper: A Deep Reinforced Model for Abstractive Summarization