AI agents could accelerate scientific discovery by automating hypothesis formation, experiment design, coding, execution, and analysis, yet existing benchmarks probe narrow skills in simplified settings. To address this gap, we introduce InnovatorBench, a benchmark-platform pair for realistic, end-to-end assessment of agents performing Large Language Model (LLM) research. It comprises 20 tasks spanning Data Construction, Filtering, Augmentation, Loss Design, Reward Design, and Scaffold Construction, which require runnable artifacts and assessment of correctness, performance, output quality, and uncertainty. To support agent operation, we develop ResearchGym, a research environment offering rich action spaces, distributed and long-horizon execution, asynchronous monitoring, and snapshot saving. We also implement a lightweight ReAct agent that couples explicit reasoning with executable planning using frontier models such as Claude-4, GPT-5, GLM-4.5, and Kimi-K2. Our experiments demonstrate that while frontier models show promise in code-driven research tasks, they struggle with fragile algorithm-related tasks and long-horizon decision making, such as impatience, poor resource management, and overreliance on template-based reasoning. Furthermore, agents require over 11 hours to achieve their best performance on InnovatorBench, underscoring the benchmark's difficulty and showing the potential of InnovatorBench to be the next generation of code-based research benchmark.

Efficient large-scale retrieval requires representations that are both compact and discriminative. Foundation models provide powerful visual and multimodal embeddings, but nearest neighbor search in these high-dimensional spaces is computationally expensive. Hashing offers an efficient alternative by enabling fast Hamming distance search with binary codes, yet existing approaches often rely on complex pipelines, multi-term objectives, designs specialized for a single learning paradigm, and long training times. We introduce CroVCA (Cross-View Code Alignment), a simple and unified principle for learning binary codes that remain consistent across semantically aligned views. A single binary cross-entropy loss enforces alignment, while coding-rate maximization serves as an anti-collapse regularizer to promote balanced and diverse codes. To implement this, we design HashCoder, a lightweight MLP hashing network with a final batch normalization layer to enforce balanced codes. HashCoder can be used as a probing head on frozen embeddings or to adapt encoders efficiently via LoRA fine-tuning. Across benchmarks, CroVCA achieves state-of-the-art results in just 5 training epochs. At 16 bits, it particularly well-for instance, unsupervised hashing on COCO completes in under 2 minutes and supervised hashing on ImageNet100 in about 3 minutes on a single GPU. These results highlight CroVCA's efficiency, adaptability, and broad applicability.

We introduce Kimi Linear, a hybrid linear attention architecture that, for the first time, outperforms full attention under fair comparisons across various scenarios -- including short-context, long-context, and reinforcement learning (RL) scaling regimes. At its core lies Kimi Delta Attention (KDA), an expressive linear attention module that extends Gated DeltaNet with a finer-grained gating mechanism, enabling more effective use of limited finite-state RNN memory. Our bespoke chunkwise algorithm achieves high hardware efficiency through a specialized variant of the Diagonal-Plus-Low-Rank (DPLR) transition matrices, which substantially reduces computation compared to the general DPLR formulation while remaining more consistent with the classical delta rule. We pretrain a Kimi Linear model with 3B activated parameters and 48B total parameters, based on a layerwise hybrid of KDA and Multi-Head Latent Attention (MLA). Our experiments show that with an identical training recipe, Kimi Linear outperforms full MLA with a sizeable margin across all evaluated tasks, while reducing KV cache usage by up to 75% and achieving up to 6 times decoding throughput for a 1M context. These results demonstrate that Kimi Linear can be a drop-in replacement for full attention architectures with superior performance and efficiency, including tasks with longer input and output lengths. To support further research, we open-source the KDA kernel and vLLM implementations, and release the pre-trained and instruction-tuned model checkpoints.

Multi-page visual documents such as manuals, brochures, presentations, and posters convey key information through layout, colors, icons, and cross-slide references. While large language models (LLMs) offer opportunities in document understanding, current systems struggle with complex, multi-page visual documents, particularly in fine-grained reasoning over elements and pages. We introduce SlideAgent, a versatile agentic framework for understanding multi-modal, multi-page, and multi-layout documents, especially slide decks. SlideAgent employs specialized agents and decomposes reasoning into three specialized levels-global, page, and element-to construct a structured, query-agnostic representation that captures both overarching themes and detailed visual or textual cues. During inference, SlideAgent selectively activates specialized agents for multi-level reasoning and integrates their outputs into coherent, context-aware answers. Extensive experiments show that SlideAgent achieves significant improvement over both proprietary (+7.9 overall) and open-source models (+9.8 overall).

Diacritics restoration in Hebrew is a fundamental task for ensuring accurate word pronunciation and disambiguating textual meaning. Despite the language's high degree of ambiguity when unvocalized, recent machine learning approaches have significantly advanced performance on this task. In this work, we present DIVRIT, a novel system for Hebrew diacritization that frames the task as a zero-shot classification problem. Our approach operates at the word level, selecting the most appropriate diacritization pattern for each undiacritized word from a dynamically generated candidate set, conditioned on the surrounding textual context. A key innovation of DIVRIT is its use of a Hebrew Visual Language Model, which processes undiacritized text as an image, allowing diacritic information to be embedded directly within the input's vector representation. Through a comprehensive evaluation across various configurations, we demonstrate that the system effectively performs diacritization without relying on complex, explicit linguistic analysis. Notably, in an ``oracle'' setting where the correct diacritized form is guaranteed to be among the provided candidates, DIVRIT achieves a high level of accuracy. Furthermore, strategic architectural enhancements and optimized training methodologies yield significant improvements in the system's overall generalization capabilities. These findings highlight the promising potential of visual representations for accurate and automated Hebrew diacritization.

Object-context shortcuts remain a persistent challenge in vision-language models, undermining zero-shot reliability when test-time scenes differ from familiar training co-occurrences. We recast this issue as a causal inference problem and ask: Would the prediction remain if the object appeared in a different environment? To answer this at inference time, we estimate object and background expectations within CLIP's representation space, and synthesize counterfactual embeddings by recombining object features with diverse alternative contexts sampled from external datasets, batch neighbors, or text-derived descriptions. By estimating the Total Direct Effect and simulating intervention, we further subtract background-only activation, preserving beneficial object-context interactions while mitigating hallucinated scores. Without retraining or prompt design, our method substantially improves both worst-group and average accuracy on context-sensitive benchmarks, establishing a new zero-shot state of the art. Beyond performance, our framework provides a lightweight representation-level counterfactual approach, offering a practical causal avenue for debiased and reliable multimodal reasoning.

Large language models (LLMs) have increasingly been applied to automatic programming code generation. This task can be viewed as a language generation task that bridges natural language, human knowledge, and programming logic. However, it remains underexplored in domains that require interaction with hardware devices, such as quantum programming, where human coders write Python code that is executed on a quantum computer. To address this gap, we introduce QCoder Benchmark, an evaluation framework that assesses LLMs on quantum programming with feedback from simulated hardware devices. Our benchmark offers two key features. First, it supports evaluation using a quantum simulator environment beyond conventional Python execution, allowing feedback of domain-specific metrics such as circuit depth, execution time, and error classification, which can be used to guide better generation. Second, it incorporates human-written code submissions collected from real programming contests, enabling both quantitative comparisons and qualitative analyses of LLM outputs against human-written codes. Our experiments reveal that even advanced models like GPT-4o achieve only around 18.97% accuracy, highlighting the difficulty of the benchmark. In contrast, reasoning-based models such as o3 reach up to 78% accuracy, outperforming averaged success rates of human-written codes (39.98%). We release the QCoder Benchmark dataset and public evaluation API to support further research. (Codes and datasets are available at https://qcoder-bench.github.io/ )

Recent efforts leverage knowledge distillation techniques to develop lightweight and practical sentiment analysis models. These methods are grounded in human-written instructions and large-scale user texts. Despite the promising results, two key challenges remain: (1) manually written instructions are limited in diversity and quantity, making them insufficient to ensure comprehensive coverage of distilled knowledge; (2) large-scale user texts incur high computational cost, hindering the practicality of these methods. To this end, we introduce CompEffDist, a comprehensive and efficient distillation framework for sentiment analysis. Our framework consists of two key modules: attribute-based automatic instruction construction and difficulty-based data filtering, which correspondingly tackle the aforementioned challenges. Applying our method across multiple model series (Llama-3, Qwen-3, and Gemma-3), we enable 3B student models to match the performance of 20x larger teacher models on most tasks. In addition, our approach greatly outperforms baseline methods in data efficiency, attaining the same performance level with only 10% of the data.

Large Language Models (LLMs) increasingly rely on external tools to perform complex, realistic tasks, yet their ability to utilize the rapidly expanding Model Contextual Protocol (MCP) ecosystem remains limited. Existing MCP research covers few servers, depends on costly manual curation, and lacks training support, hindering progress toward real-world deployment. To overcome these limitations, we introduce MCP-Flow, an automated web-agent-driven pipeline for large-scale server discovery, data synthesis, and model training. MCP-Flow collects and filters data from 1166 servers and 11536 tools, producing 68733 high-quality instruction-function call pairs and 6439 trajectories, far exceeding prior work in scale and diversity. Extensive experiments demonstrate MCP-Flow's effectiveness in driving superior MCP tool selection, function-call generation, and enhanced agentic task performance. MCP-Flow thus provides a scalable foundation for advancing LLM agents' proficiency in real-world MCP environments. MCP-Flow is publicly available at \href{https://github.com/wwh0411/MCP-Flow}{https://github.com/wwh0411/MCP-Flow}.

Large Language Models (LLMs) often produce explanations that do not faithfully reflect the factors driving their predictions. In healthcare settings, such unfaithfulness is especially problematic: explanations that omit salient clinical cues or mask spurious shortcuts can undermine clinician trust and lead to unsafe decision support. We study how inference and training-time choices shape explanation faithfulness, focusing on factors practitioners can control at deployment. We evaluate three LLMs (GPT-4.1-mini, LLaMA 70B, LLaMA 8B) on two datasets-BBQ (social bias) and MedQA (medical licensing questions), and manipulate the number and type of few-shot examples, prompting strategies, and training procedure. Our results show: (i) both the quantity and quality of few-shot examples significantly impact model faithfulness; (ii) faithfulness is sensitive to prompting design; (iii) the instruction-tuning phase improves measured faithfulness on MedQA. These findings offer insights into strategies for enhancing the interpretability and trustworthiness of LLMs in sensitive domains.