Large Language Models (LLMs) have rapidly advanced in recent years, scaling up in both parameter count and context length. However, as context windows extend from thousands to hundreds of thousands of tokens, attention computation becomes the dominant source of memory usage and runtime in decoding stages, severely limiting the efficiency and scalability of long-context LLMs. Sparse attention has emerged as a promising solution, reducing complexity by computing attention over only a subset of context tokens. However, the sparse attention for Multi-head Latent Attention(MLA) which is a variant of standard MHA is rarely studied. In this paper, we introduce RoPE-based Blockwise Sparse Attention (RoBSA), a method designed specifically for MLA during the decoding stage of model inference. RoBSA leverages the decoupled nature of RoPE within MLA to implement token selection in a blockwise manner. RoBSA is a lightweight, training-free, and layer-aware algorithm that can be integrated in a plug-and-play fashion. Our method significantly reduces end-to-end inference latency in the decoding stage by up to 2.55x with minimal accuracy loss compared to full attention in long-context scenarios for very large models.

Despite the potential of multi-turn self-reflection to improve LLM reasoning, its effectiveness in practice is severely constrained by a failure mode we term the Echo Trap.Specifically, this phenomenon gives rise to two coupled problems: (1) the model becomes limited by its inherent capabilities and tends to repeat earlier reflections to preserve reward signals; (2) once such “copy” behavior is reinforced, the model ceases to try new strategies, leading to exploration collapse.We attribute this issue to imprecise credit assignment during training, as standard GRPO assigns rewards at the trajectory level, making it difficult to distinguish which reflection steps contribute to improved outcomes.To address this limitation, we propose a tree-structured extension of GRPO for multi-turn self-reflection, which enables more accurate advantage estimation.Through extensive experiments, we analyze the Echo Trap and demonstrate that our method effectively mitigates behavior collapse and improves performance across multiple benchmarks.

Deep search agents that combine large language models with retrieval tools excel at complex, multi-hop queries. Yet, existing benchmarks such as BrowseComp rely on black-box web search APIs, facing key limitations. (1) Fairness: for agents, dynamic and opaque web APIs hinder reproducibility and fair comparisons across agents. (2) Disentanglement: for retrieval, the lack of a fixed document corpus makes it impossible to isolate retriever contributions from end-to-end search agent accuracy. We introduce BrowseComp-Plus, a benchmark derived from BrowseComp that employs a fixed, human-verified corpus, enabling controlled retrieval for deep search agents. BrowseComp-Plus clearly distinguishes agent performance: with a BM25 retriever, the open-source Search-R1 achieves 3.86% accuracy, while GPT-5 achieves 55.9%. Additionally, BrowseComp-Plus makes retrieval gains explicit: pairing GPT-5 with Qwen3-Embedding-8B retriever further improves accuracy to 70.1% while reducing search calls. Overall, BrowseComp-Plus provides a fair and disentangled testbed, advancing both deep search agent evaluation and retrieval research for agentic search. Code and data can be found at: https://texttron.github.io/BrowseComp-Plus/

Improving the mathematical reasoning capabilities of Large Language Models (LLMs) is critical for advancing artificial intelligence. However, access to extensive, diverse, and high-quality reasoning datasets remains a significant challenge, particularly for the open-source community. In this paper, we propose ScaleQuest, a novel, scalable, and cost-effective data synthesis method that enables the generation of large-scale mathematical reasoning datasets using lightweight 7B-scale models. ScaleQuest introduces a two-stage question-tuning process comprising Question Fine-Tuning (QFT) and Question Preference Optimization (QPO) to unlock the question generation capabilities of problem-solving models. By generating diverse questions from scratch – without relying on powerful proprietary models or seed data – we produce a dataset of 1 million problem-solution pairs. Our experiments demonstrate that models trained on our data outperform existing open-source datasets in both in-domain and out-of-domain evaluations. Furthermore, our approach shows continued performance improvement as the volume of training data increases, highlighting its potential for ongoing data scaling. The extensive improvements observed in code reasoning tasks demonstrate the generalization capabilities of our proposed method. Our work provides the open-source community with a practical solution to enhance the mathematical reasoning abilities of LLMs.