Recent trends in test-time scaling for reasoning models (e.g., OpenAI o1, DeepSeek-R1) have led to remarkable improvements through long Chain-of-Thought (CoT). However, existing benchmarks mainly focus on immediate, single-horizon tasks, failing to adequately evaluate models’ ability to understand and respond to complex, long-horizon scenarios. To address this incomplete evaluation of Large Reasoning Models (LRMs), we propose R-HORIZON, a method designed to stimulate long-horizon reasoning behaviors in LRMs through query composition. Based on R-HORIZON, we construct a long-horizon reasoning benchmark, comprising complex multi-step reasoning tasks with interdependent problems that span long reasoning horizons. Through comprehensive evaluation of LRMs using the R-HORIZON benchmark, we find that even the most advanced LRMs suffer significant performance degradation. Our analysis reveals that LRMs exhibit limited effective reasoning length and struggle to allocate thinking budget across multiple problems appropriately. Recognizing these limitations, we use R-HORIZON to construct long-horizon reasoning data for reinforcement learning with verified rewards (RLVR). Compared to training with single-horizon data, RLVR with R-HORIZON not only substantially improves performance on the multi-horizon reasoning tasks, but also promotes accuracy on standard reasoning tasks (+7.5 on AIME2024). These results position R-HORIZON as a scalable, controllable, and low-cost paradigm for enhancing and evaluating the long-horizon reasoning capabilities of LRMs.

Long-context language models now advertise context windows up to millions of tokens, yet evaluations typically report a single length or a narrow task family, masking two failure modes: performance can collapse as length grows, and strong retrieval need not transfer to downstream use. We present ATLAS, a benchmarking framework that redefines long-context evaluation as length-dependent capability profiling. ATLAS contributes three methodological principles:(i) a layered taxonomy separating foundational operations from application workloads so failures can be attributed, (ii) length-aware AUC scoring that integrates score-length curves over a fixed 8K-1M grid, replacing single-point metrics with full degradation profiles, and (iii) ATLAScore, a harmonic-mean aggregate over taxonomy categories that penalizes imbalanced profiles, with end-to-end uncertainty propagation from subset scores through the nonlinear final aggregate. We instantiate the framework across eight capability dimensions with nine auditable components and 6,438 instances, and evaluate 26 models. Gemini-3.1-Pro-Preview leads at 128K, Claude-Opus-4.6 leads at 1M. Rankings reshuffle substantially between ATLASscore@8K-128K and ATLASscore@8K-1M: 7 models move by at least two ranks, and the two taxonomy layers share only 61

We introduce LongCat ZigZag Attention (LoZA), which is a sparse attention scheme designed to transform any existing full-attention models into sparse versions with rather limited compute budget. In long-context scenarios, LoZA can achieve significant speed-ups both for prefill-intensive (e.g., retrieval-augmented generation) and decode-intensive (e.g., tool-integrated reasoning) cases. Specifically, by applying LoZA to LongCat-Flash during mid-training, we serve LongCat-Flash-Exp as a long-context foundation model that can swiftly process up to 1 million tokens, enabling efficient long-term reasoning and long-horizon agentic capabilities.

Processing structured tabular data, particularly large and lengthy tables, constitutes a fundamental yet challenging task for large language models (LLMs). However, existing long-context benchmarks like Needle-in-a-Haystack primarily focus on unstructured text, neglecting the challenge of diverse structured tables. Meanwhile, previous tabular benchmarks mainly consider downstream tasks that require high-level reasoning abilities, and overlook models' underlying fine-grained perception of individual table cells, which is crucial for practical and robust LLM-based table applications. To address this gap, we introduce \textsc{NeedleInATable} (NIAT), a new long-context tabular benchmark that treats each table cell as a ``needle'' and requires models to extract the target cell based on cell locations or lookup questions. Our comprehensive evaluation of various LLMs and multimodal LLMs reveals a substantial performance gap between popular downstream tabular tasks and the simpler NIAT task, suggesting that they may rely on dataset-specific correlations or shortcuts to obtain better benchmark results but lack truly robust long-context understanding towards structured tables. Furthermore, we demonstrate that using synthesized NIAT training data can effectively improve performance on both NIAT task and downstream tabular tasks, which validates the importance of NIAT capability for LLMs' genuine table understanding ability. Our data, code and models will be released to facilitate future research.

We present LongCat-Flash-Thinking, an efficient 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model. Its advanced capabilities are cultivated through a meticulously crafted training process, beginning with long Chain-of-Thought (CoT) data cold-start and culminating in large-scale Reinforcement Learning (RL). We first employ a well-designed cold-start training strategy, which significantly enhances the reasoning potential and equips the model with specialized skills in both formal and agentic reasoning. Then, a core innovation is our domain-parallel training scheme, which decouples optimization across distinct domains (e.g., STEM, Code, Agentic) and subsequently fuses the resulting expert models into a single, nearly Pareto-optimal model. This entire process is powered by our Dynamic ORchestration for Asynchronous rollout (DORA) system, a large-scale RL framework that delivers a greater than threefold training speedup over synchronous methods on tens of thousands of accelerators. As a result, LongCat-Flash-Thinking achieves state-of-the-art performance among open-source models on a suite of complex reasoning tasks. The model exhibits exceptional efficiency in agentic reasoning, reducing average token consumption by 64.5

Large Language Models (LLMs) are increasingly prevalent in the field of long-context modeling, however, their inference computational costs have become a critical bottleneck hindering the advancement of tasks such as agents and multimodal applications. This report conducts a preliminary investigation into the effectiveness and theoretical mechanisms of the Top-k Attention mechanism during both the decoding and training phases. First, we validate the effectiveness of exact Top-k Decoding through extensive experimentation. Experiments demonstrate that retaining only the pivotal Keys with the highest similarity to the Query as the context window during the decoding stage achieves performance comparable to, or even surpassing, full attention on downstream tasks such as HELMET and LongBench v2. Second, we further explore the native Top-k Attention training strategy. Experiments confirm that ensuring the consistency between training and inference regarding Top-k Attention operations facilitates the further unlocking of Top-k Decoding's potential, thereby significantly enhancing model performance. Furthermore, considering the high computational complexity of exact Top-k Attention, we investigate the impact of approximate Top-k algorithm precision on downstream tasks. Our research confirms a positive correlation between downstream task performance and approximation fidelity, and we provide statistical evaluations of the Lightning Indexer's precision within the DeepSeek-V3.2-Exp model. Finally, this report provides a theoretical interpretation from the perspective of Entropy. Experimental observations indicate that models subjected to Top-k Attention SFT exhibit a distinct phenomenon of entropy reduction in downstream tasks, which validates the hypothesis that low-entropy states are better adapted to Top-k Decoding.

The quadratic computational complexity of the attention mechanism in current Large Language Models (LLMs) renders inference with long contexts prohibitively expensive. To address this challenge, various approaches aim to retain critical portions of the context to optimally approximate Full Attention (FA) through Key-Value (KV) compression or Sparse Attention (SA), enabling the processing of virtually unlimited text lengths in a streaming manner. However, these methods struggle to achieve performance levels comparable to FA, particularly in retrieval tasks. In this paper, our analysis of attention head patterns reveals that LLMs' attention distributions show strong local correlations, naturally reflecting a chunking mechanism for input context. We propose Ltri-LLM framework, which divides KVs into spans, stores them in an offline index, and retrieves the relevant KVs into memory for various queries. Experimental results on popular long text benchmarks show that Ltri-LLM can achieve performance close to FA while maintaining efficient, streaming-based inference.

Most previous sequential labeling models are task-specific, while recent years have witnessed the rise of generative models due to the advantage of unifying all named entity recognition (NER) tasks into the encoder-decoder framework. Although achieving promising performance, our pilot studies demonstrate that existing generative models are ineffective at detecting entity boundaries and estimating entity types. In this paper, we propose a multi-task Transformer, which incorporates an entity boundary detection task into the named entity recognition task. More concretely, we achieve entity boundary detection by classifying the relations between tokens within the sentence. To improve the accuracy of entity-type mapping during decoding, we adopt an external knowledge base to calculate the prior entity-type distributions and then incorporate the information into the model via the self- and cross-attention mechanisms. We perform experiments on extensive NER benchmarks, including flat, nested, and discontinuous NER datasets involving long entities. It substantially increases nearly +0.3 similar to +1.5 F-1 scores across a broad spectrum or performs closely to the best generative NER model. Experimental results show that our approach improves the performance of the generative NER model considerably.

Currently, the reduction in the parameter scale of large-scale pre-trained language models (PLMs) through knowledge distillation has greatly facilitated their widespread deployment on various devices. However, the deployment of knowledge distillation systems faces great challenges in real-world industrial-strength applications, which require the use of complex distillation methods on even larger-scale PLMs (over 10B), limited by memory on GPUs and the switching of methods. To overcome these challenges, we propose GKD, a general knowledge distillation framework that supports distillation on larger-scale PLMs using various distillation methods. With GKD, developers can build larger distillation models on memory-limited GPUs and easily switch and combine different distillation methods within a single framework. Experimental results show that GKD can support the distillation of at least 100B-scale PLMs and 25 mainstream methods on 8 NVIDIA A100 (40GB) GPUs.

Named entity recognition (NER) is an important research problem in natural language processing. There are three types of NER tasks, including flat, nested and discontinuous entity recognition. Most previous sequential labeling models are task-specific, while recent years have witnessed the rising of generative models due to the advantage of unifying all NER tasks into the seq2seq model framework. Although achieving promising performance, our pilot studies demonstrate that existing generative models are ineffective at detecting entity boundaries and estimating entity types. This paper proposes a multi-task Transformer, which incorporates an entity boundary detection task into the named entity recognition task. More concretely, we achieve entity boundary detection by classifying the relations between tokens within the sentence. To improve the accuracy of entity-type mapping during decoding, we adopt an external knowledge base to calculate the prior entity-type distributions and then incorporate the information into the model via the self and cross-attention mechanisms. We perform experiments on an extensive set of NER benchmarks, including two flat, three nested, and three discontinuous NER datasets. Experimental results show that our approach considerably improves the generative NER model's performance.

Neural models with an encoder-decoder framework provide a feasible solution to Question Generation (QG). However, after analyzing the model vocabulary we find that current models (both RNN-based and pre-training based) have more than 23\% inflected forms. As a result, the encoder will generate separate embeddings for the inflected forms, leading to a waste of training data and parameters. Even worse, in decoding these models are vulnerable to irrelevant noise and they suffer from high computational costs. In this paper, we propose an approach to enhance the performance of QG by fusing word transformation. Firstly, we identify the inflected forms of words from the input of encoder, and replace them with the root words, letting the encoder pay more attention to the repetitive root words. Secondly, we propose to adapt QG as a combination of the following actions in the encode-decoder framework: generating a question word, copying a word from the source sequence or generating a word transformation type. Such extension can greatly decrease the size of predicted words in the decoder as well as noise. We apply our approach to a typical RNN-based model and \textsc{UniLM} to get the improved versions. We conduct extensive experiments on SQuAD and MS MARCO datasets. The experimental results show that the improved versions can significantly outperform the corresponding baselines in terms of BLEU, ROUGE-L and METEOR as well as time cost.

Pre-trained Language Models (PLMs) have achieved remarkable performance gains across numerous downstream tasks in natural language understanding. Various Chinese PLMs have been successively proposed for learning better Chinese language representation. However, most current models use Chinese characters as inputs and are not able to encode semantic information contained in Chinese words. While recent pre-trained models incorporate both words and characters simultaneously, they usually suffer from deficient semantic interactions and fail to capture the semantic relation between words and characters. To address the above issues, we propose a simple yet effective PLM CLOWER, which adopts the Contrastive Learning Over Word and charactER representations. In particular, CLOWER implicitly encodes the coarse-grained information (i.e., words) into the fine-grained representations (i.e., characters) through contrastive learning on multi-grained information. CLOWER is of great value in realistic scenarios since it can be easily incorporated into any existing fine-grained based PLMs without modifying the production pipelines. Extensive experiments conducted on a range of downstream tasks demonstrate the superior performance of CLOWER over several state-of-the-art baselines.

Text matching is a fundamental research problem in natural language understanding. Interaction-based approaches treat the text pair as a single sequence and encode it through cross encoders, while representation-based models encode the text pair independently with siamese or dual encoders. Interaction-based models require dense computations and thus are impractical in real-world applications. Representation-based models have become the mainstream paradigm for efficient text matching. However, these models suffer from severe performance degradation due to the lack of interactions between the pair of texts. To remedy this, we propose a Virtual InteRacTion mechanism (VIRT) for improving representation-based text matching while maintaining its efficiency. In particular, we introduce an interactive knowledge distillation module that is only applied during training. It enables deep interaction between texts by effectively transferring knowledge from the interaction-based model. A light interaction strategy is designed to fully leverage the learned interactive knowledge. Experimental results on six text matching benchmarks demonstrate the superior performance of our method over several state-of-the-art representation-based models. We further show that VIRT can be integrated into existing methods as plugins to lift their performances.

Recently, the retrieval models based on dense representations have been gradually applied in the first stage of the document retrieval tasks, showing better performance than traditional sparse vector space models. To obtain high efficiency, the basic structure of these models is Bi-encoder in most cases. However, this simple structure may cause serious information loss during the encoding of documents since the queries are agnostic. To address this problem, we design a method to mimic the queries on each of the documents by an iterative clustering process and represent the documents by multiple pseudo queries (i.e., the cluster centroids). To boost the retrieval process using approximate nearest neighbor search library, we also optimize the matching function with a two-step score calculation procedure. Experimental results on several popular ranking and QA datasets show that our model can achieve state-of-the-art results.

In this paper, we focus on the problem of keyword and document matching by considering different relevance levels. In our recommendation system, different people follow different hot keywords with interest. We need to attach documents to each keyword and then distribute the documents to people who follow these keywords. The ideal documents should have the same topic with the keyword, which we call topic-aware relevance. In other words, topic-aware relevance documents are better than partially-relevance ones in this application. However, previous tasks never define topic-aware relevance clearly. To tackle this problem, we define a three-level relevance in keyword-document matching task: topic-aware relevance, partially-relevance and irrelevance. To capture the relevance between the short keyword and the document at above-mentioned three levels, we should not only combine the latent topic of the document with its deep neural representation, but also model complex interactions between the keyword and the document. To this end, we propose a Two-stage Interaction and Topic-Aware text matching model (TITA). In terms of “topic-aware”, we introduce neural topic model to analyze the topic of the document and then use it to further encode the document. In terms of “two-stage interaction”, we propose two successive stages to model complex interactions between the keyword and the document. Extensive experiments reveal that TITA outperforms other well-designed baselines and shows excellent performance in our recommendation system.

Recently, a zero-shot entity linking task is introduced to challenge the generalization ability of entity linking models. In this task, mentions must be linked to unseen entities and only the textual information is available. In order to make full use of the documents, previous work has proposed a BERT-based model which can only take fixed length of text as input. However, the key information for entity linking may exist in nearly everywhere of the documents thus the proposed model cannot capture them all. To leverage more textual information and enhance text understanding capability, we propose a bidirectional multi-paragraph reading model for the zero-shot entity linking task. Firstly, the model treats the mention context as a query and matches it with multiple paragraphs of the entity description documents. Then, the mention-aware entity representation obtained from the first step is used as a query to match multiple paragraphs in the document containing the mention through an entity-mention attention mechanism. In particular, a new pre-training strategy is employed to strengthen the representative ability. Experimental results show that our bidirectional model can capture long-range context dependencies and outperform the baseline model by 3-4% in terms of accuracy.

In this paper, we propose a new ranking model DR-BERT, which improves the Document Retrieval (DR) task by a task-adaptive training process and a Segmented Token Recovery Mechanism (STRM). In the task-adaptive training, we first pre-train DR-BERT to be domain-adaptive and then make the two-phase fine-tuning. In the first-phase fine-tuning, the model learns query-document matching patterns regarding different query types in a pointwise way. Next, in the second-phase fine-tuning, the model learns document-level ranking features and ranks documents with regard to a given query in a listwise manner. Such pointwise plus listwise fine-tuning enables the model to minimize errors in the document ranking by incorporating ranking-specific supervisions. Meanwhile, the model derived from pointwise fine-tuning is also used to reduce noise in the training data of the listwise fine-tuning. On the other hand, we present STRM which can compute OOV word representation and contextualization more precisely in BERT-based models. As an effective strategy in DR-BERT, STRM improves the matching perfromance of OOV words between a query and a document. Notably, our DR-BERT model keeps in the top three on the MS MARCO leaderboard since May 20, 2020.

With the booming of pre-trained transformers, representation-based models based on Siamese transformer encoders have become mainstream techniques for efficient text matching. However, these models suffer from severe performance degradation due to the lack of interaction between the text pair, compared with interaction-based models. Prior arts attempt to address this through performing extra interaction for Siamese encoded representations, while the interaction during encoding is still ignored. To remedy this, we propose a Virtual InteRacTion mechanism (VIRT) to transfer interactive knowledge from interaction-based models into Siamese encoders through attention map distillation. As a train-time-only component, VIRT could completely maintain the high efficiency of the Siamese structure and brings no extra computation cost during inference. To fully utilize the learned interactive knowledge, we further design a VIRT-adapted interaction strategy. Experimental results on multiple text matching datasets demonstrate that our method outperforms state-of-the-art representation-based models. What's more, VIRT can be easily integrated into existing representation-based methods to achieve further improvements.