Code has emerged as a precise and executable medium for reasoning and action in the agent era. Yet, progress has largely focused on language-centric tasks such as program synthesis and debugging, leaving visual-centric coding underexplored. Inspired by how humans reason over sketches, we advocate SVG code as a compact, interpretable, and executable visual representation. We introduce VCode, a benchmark that reframes multimodal understanding as code generation: given an image, a model must produce SVG that preserves symbolic meaning for downstream reasoning. VCode covers three domains - general commonsense (MM-Vet), professional disciplines (MMMU), and visual-centric perception (CV-Bench). To assess symbolic fidelity, we propose CodeVQA, a novel evaluation protocol in which a policy model answers questions over rendered SVGs; correct answers indicate faithful symbolic preservation. Empirically, frontier VLMs struggle to generate faithful SVGs, revealing a persistent gap between language-centric and visual-centric coding. To close this gap, we introduce VCoder, an agentic framework that augments VLMs along two axes: (i) Thinking with Revision, which iteratively analyzes discrepancies and refines SVG code; and (ii) Acting with Visual Tools, where detectors and parsers supply structured cues such as objects, shapes, and text beyond the model's intrinsic capacity. Across benchmarks, frontier VLMs with strong reasoning capabilities score well overall yet remain limited in professional knowledge and 3D reasoning. VCoder delivers a 12.3-point overall gain over the top-performing Claude-4-Opus. Human studies show that both humans and VLMs perform worse on rendered SVGs, their consistency reveals the promise of symbolic visual representation. The benchmark and code are available at https://github.com/CSU-JPG/VCode.

Most text retrievers generate one query vector to retrieve relevant documents. Y et, the conditional distribution of relevant documents for the query may be multi-modal, e.g., representing different interpretations of the query. We first quantify the limitations of existing retrievers. All retrievers we evaluate struggle more as the distance between target document embeddings grows. Our model autoregressively generates multiple query vectors, and all the predicted query vectors are used to retrieve documents from the corpus. We show that on the synthetic vectorized data, the proposed method could capture multiple target distributions perfectly, showing 4x better performance than single embedding model. We also fine-tune our model on real-world multi-answer retrieval datasets and evaluate in-domain. AMER presents 4 and 21% relative gains over single-embedding baselines on two datasets we evaluate on. Furthermore, we consistently observe larger gains on the subset of dataset where the embeddings of the target documents are less similar to each other. We demonstrate the potential of using a multi-query vector retriever and open up a new direction for future work. As large language models (LLMs) have limited, out-dated parametric knowledge, augmenting knowledge at inference time by prepending retrieved documents has risen as a de facto solution (Fan et al., 2024; Gao et al., 2023). Recovering a diverse set of documents is crucial to provide comprehensive information (Xu et al., 2023), as an answer providing partial information can be technically correct yet misleading to users. In this work, we study retrieving a diverse set of documents per query. We first analyze the behaviors of existing retrievers (Izacard et al., 2022; Y ang et al., 2025b; Zhang et al., 2025; Lee et al., 2025a) on datasets (Min et al., 2020; Amouyal et al., 2023) containing questions that admit multiple valid answers.

The chemical space of drug-like molecules is vast, motivating the development of generative models that must learn broad chemical distributions, enable conditional generation by capturing structure-property representations, and provide fast molecular generation. Meeting the objectives depends on modeling choices, including the probabilistic modeling approach, the conditional generative formulation, the architecture, and the molecular input representation. To address the challenges, we present STAR-VAE (Selfies-encoded, Transformer-based, AutoRegressive Variational Auto Encoder), a scalable latent-variable framework with a Transformer encoder and an autoregressive Transformer decoder. It is trained on 79 million drug-like molecules from PubChem, using SELFIES to guarantee syntactic validity. The latent-variable formulation enables conditional generation: a property predictor supplies a conditioning signal that is applied consistently to the latent prior, the inference network, and the decoder. Our contributions are: (i) a Transformer-based latent-variable encoder-decoder model trained on SELFIES representations; (ii) a principled conditional latent-variable formulation for property-guided generation; and (iii) efficient finetuning with low-rank adapters (LoRA) in both encoder and decoder, enabling fast adaptation with limited property and activity data. On the GuacaMol and MOSES benchmarks, our approach matches or exceeds baselines, and latent-space analyses reveal smooth, semantically structured representations that support both unconditional exploration and property-aware generation. On the Tartarus benchmarks, the conditional model shifts docking-score distributions toward stronger predicted binding. These results suggest that a modernized, scale-appropriate VAE remains competitive for molecular generation when paired with principled conditioning and parameter-efficient finetuning.

Large language models (LLMs) exhibit complementary strengths across domains and come with varying inference costs, motivating the design of multi-agent LLM systems where specialized models collaborate efficiently. Existing approaches predominantly rely on decentralized frameworks, which invoke multiple LLMs for every input and thus lead to substantial and uncontrolled inference costs. In this work, we introduce a centralized multi-LLM framework, where a controller LLM selectively coordinates a pool of expert models in a cost-efficient and cost-controllable manner. We formulate this coordination problem as reinforcement learning with dual objectives: maximizing task performance while minimizing the overall inference cost. In addition, we expect the multi-agent system to have adapted behavior with different budget conditions during inference. To this end, we propose CoRL, a reinforcement learning framework that optimizes the performance cost trade-off in a controllable multi-budget setting. Experiments on four diverse benchmarks demonstrate that CoRL enables a single system to surpass the best expert LLM under high-budget settings, while maintaining strong performance in more economical low-budget modes, highlighting the effectiveness of centralized coordination for scalable and cost-efficient multi-agent LLM systems.

Classical probabilistic graphical models face fundamental challenges in modern data environments, which are characterized by high dimensionality, source heterogeneity, and stringent data-sharing constraints. In this work, we revisit the Ising model, a well-established member of the Markov Random Field (MRF) family, and develop a distributed framework that enables scalable and privacy-preserving representation learning from large-scale binary data with inherent low-rank structure. Our approach optimizes a non-convex surrogate loss function via bi-factored gradient descent, offering substantial computational and communication advantages over conventional convex approaches. We evaluate our algorithm on multi-institutional electronic health record (EHR) datasets from 58,248 patients across the University of Pittsburgh Medical Center (UPMC) and Mass General Brigham (MGB), demonstrating superior performance in global representation learning and downstream clinical tasks, including relationship detection, patient phenotyping, and patient clustering. These results highlight a broader potential for statistical inference in federated, high-dimensional settings while addressing the practical challenges of data complexity and multi-institutional integration.

Clients are evolving beyond chat completion, and now include a variety of innovative inference-time scaling and deep reasoning techniques. At the same time, inference servers remain heavily optimized for chat completion. Prior work has shown that large improvements to KV cache hit rate are possible if inference servers evolve towards these non-chat use cases. However, they offer solutions that are also optimized for a single use case, RAG. In this paper, we introduce the span query to generalize the interface to the inference server. We demonstrate that chat, RAG, inference-time scaling, and agentic workloads can all be expressed as span queries. We show how the critical distinction that had been assumed by prior work lies in whether the order of the inputs matter -- do they commute? In chat, they do not. In RAG, they often do. This paper introduces span queries, which are expression trees of inference calls, linked together with commutativity constraints. We describe span query syntax and semantics. We show how they can be automatically optimized to improve KV cache locality. We show how a small change to vLLM (affecting only 492 lines) can enable high-performance execution of span queries. Using this stack, we demonstrate that span queries can achieve 10-20x reductions in TTFT for two distinct non-chat use cases. Finally, we show that span queries can also be optimized to improve attention locality, so as to avoid the so-called lost-in-the-middle problem. We demonstrate that an attention-optimized span query on a 2b parameter model vastly outperforms the accuracy of a stock inference server using an 8b model.

Current evaluations of Large Language Model (LLM) agents primarily emphasize task completion, often overlooking resource efficiency and adaptability. This neglects a crucial capability: agents' ability to devise and adjust cost-optimal plans in response to changing environments. To bridge this gap, we introduce CostBench, a scalable, cost-centric benchmark designed to evaluate agents' economic reasoning and replanning abilities. Situated in the travel-planning domain, CostBench comprises tasks solvable via multiple sequences of atomic and composite tools with diverse, customizable costs. It also supports four types of dynamic blocking events, such as tool failures and cost changes, to simulate real-world unpredictability and necessitate agents to adapt in real time. Evaluating leading open-sourced and proprietary models on CostBench reveals a substantial gap in cost-aware planning: agents frequently fail to identify cost-optimal solutions in static settings, with even GPT-5 achieving less than 75% exact match rate on the hardest tasks, and performance further dropping by around 40% under dynamic conditions. By diagnosing these weaknesses, CostBench lays the groundwork for developing future agents that are both economically rational and robust.

Convolutional neural networks (CNNs) underpin many modern computer vision systems. With applications ranging from common to critical areas, a need to explain and understand the model and its decisions (XAI) emerged. Prior works suggest that in the top layers of CNNs, the individual channels can be attributed to classifying human-understandable concepts. Concept relevance propagation (CRP) methods can backtrack predictions to these channels and find images that most activate these channels. However, current CRP workflows are largely manual: experts must inspect activation images to name the discovered concepts and must synthesize verbose explanations from relevance maps, limiting the accessibility of the explanations and their scalability. To address these issues, we introduce Large Language model EXplaIns COncept Relevance Propagation (LLEXICORP), a modular pipeline that couples CRP with a multimodal large language model. Our approach automatically assigns descriptive names to concept prototypes and generates natural-language explanations that translate quantitative relevance distributions into intuitive narratives. To ensure faithfulness, we craft prompts that teach the language model the semantics of CRP through examples and enforce a separation between naming and explanation tasks. The resulting text can be tailored to different audiences, offering low-level technical descriptions for experts and high-level summaries for non-technical stakeholders. We qualitatively evaluate our method on various images from ImageNet on a VGG16 model. Our findings suggest that integrating concept-based attribution methods with large language models can significantly lower the barrier to interpreting deep neural networks, paving the way for more transparent AI systems.

The trajectory of AI development suggests that we will increasingly rely on agent-based systems composed of independently developed agents with different information, privileges, and tools. The success of these systems will critically depend on effective collaboration among these heterogeneous agents, even under partial observability. Despite intense interest, few empirical studies have evaluated such agent-agent collaboration at scale. We propose a collaborative maze-solving benchmark that (i) isolates collaborative capabilities, (ii) modulates problem complexity, (iii) enables scalable automated grading, and (iv) imposes no output-format constraints, preserving ecological plausibility. Using this framework, we evaluate 32 leading open- and closed-source models in solo, homogeneous, and heterogeneous pairings. Our results reveal a "collaboration gap": models that perform well solo often degrade substantially when required to collaborate. Collaboration can break down dramatically; for instance, small distilled models that solve mazes well alone may fail almost completely in certain pairings. We find that starting with the stronger agent often improves outcomes, motivating a "relay inference" approach where the stronger agent leads before handing off to the weaker one, closing much of the gap. Our findings argue for (1) collaboration-aware evaluation, (2) training strategies developed to enhance collaborative capabilities, and (3) interaction design that reliably elicits agents' latent skills, guidance that applies to AI-AI and human-AI collaboration.

Large Language Models (LLMs) achieve remarkable reasoning capabilities through transformer architectures with attention mechanisms. However, transformers suffer from quadratic time and memory complexity in the attention module (MHA) and require caching key-value states during inference, which severely limits throughput and scalability. High inference throughput is critical for agentic tasks, long-context reasoning, efficient deployment under high request loads, and more efficient test-time compute scaling. State Space Models (SSMs) such as Mamba offer a promising alternative with linear inference complexity and a constant memory footprint via recurrent computation with fixed-size hidden states. In this technical report we introduce the Apriel-H1 family of hybrid LLMs that combine transformer attention and SSM sequence mixers for efficient reasoning at 15B model size. These models are obtained through incremental distillation from a pretrained reasoning transformer, Apriel-Nemotron-15B-Thinker, progressively replacing less critical attention layers with linear Mamba blocks. We release multiple post-distillation variants of Apriel-H1-15B-Thinker with different SSM-to-MHA ratios and analyse how reasoning performance degrades as more Mamba layers replace MHA. Additionally, we release a 30/50 hybrid variant of Apriel-H1, further fine-tuned on a supervised dataset of reasoning traces, achieving over 2x higher inference throughput when deployed in the production-ready vLLM environment, with minimal degradation in reasoning performance. This shows that distilled hybrid SSM-Transformer architectures can deliver substantial efficiency gains over the pretrained transformer equivalent without substantially compromising the reasoning quality.