Agentic Context Learning with Self-Discovered Specification

Context learning is an emerging inference-time task where LLMs must learn and apply novel, task-specific knowledge from intricate contexts absent from pre-training; even frontier models score under 24% task success. In this work, we conduct a comprehensive empirical study to understand why this setting remains difficult. A natural hypothesis is that failures stem from content access; yet across twelve retrieval, reflection, and verification baselines on CL-Bench, an extensive context learning benchmark, we find limited gains over direct full-context prompting. Further failure analysis reveals a key finding: unlike typical long-context tasks such as long document understanding, context learning requires not only recovering local content but also acquiring local specifications that are often unspecified in the query but distributed across the context: domain-specific formats, local rules, and completeness conditions. Across all 31,592 rubric items, we find that 55.4% clearly evaluate specification acquisition, while only 22.6% evaluate content acquisition. Moreover, despite 76.7% of specifications being unspecified in the user query, 95.5% are traceable to the context, indicating these are learnable obligations rather than hidden requirements. To validate this diagnosis, we design a deliberately simple intervention PSCI (private specification-contract induction) which extracts local specifications and enforces them through adversarial checking and repair; PSCI achieves state-of-the-art 28.14% with GPT-5.1 (+5.59 pp absolute and +24.8% relative) on CL-Bench, replicated on Qwen3.5-27B (+5.28 pp) and Gemini 3 Pro (+6.17 pp). Seventeen ablations further isolate the role of task-specific specifications. Overall, our results suggest context learning hinges on not only content acquisition but also specification acquisition.

Spinning Straw into Gold: Relabeling LLM Agent Trajectories in Hindsight for Successful Demonstrations

Large language model agents operate in partially observable, long-horizon settings where obtaining supervision remains a major bottleneck. We address this by utilizing a source of supervision overlooked in existing post-training methods: unintended yet successful goals embedded within agent rollouts. Specifically, we introduce Hindsight Supervised Learning (HSL), where an auxiliary LLM reviews each completed trajectory and relabels it with all of the natural-language goals the agent actually achieved. HSL then pairs the trajectory with its relabeled goals and uses these pairs for additional fine-tuning. To mitigate suboptimality in the relabeled data, we propose two learning techniques for HSL, irrelevant-action masking and sample reweighting. Our experiments show that HSL is flexible and compatible with existing post-training pipelines. It improves both SFT and DPO, with larger gains on long-horizon tasks with more diverse goal spaces. Moreover, HSL is sample-efficient: on ALFWorld, it surpasses baselines trained on the full dataset while using only one quarter of the ground-truth demonstrations.

DocPrune:Efficient Document Question Answering via Background, Question, and Comprehension-aware Token Pruning

Recent advances in vision-language models have demonstrated remarkable performance across diverse multi-modal tasks, including document question answering that leverages structured visual cues from text, tables, and figures. However, unlike natural images, document images contain large backgrounds and only sparse supporting evidence, leading to the inefficient consumption of substantial computational resources, especially for long documents. We observe that existing token-reduction methods for natural images and videos fall short in utilizing the structural sparsity unique to documents. To address this, we propose DocPrune, a training-free and progressive document token pruning framework designed for efficient long-document understanding. The proposed method preserves only the essential tokens for the task while removing unnecessary ones, such as background or question-irrelevant tokens. Moreover, it automatically selects the appropriate layers to initiate token pruning based on the model's level of comprehension. Our experiments on the M3DocRAG show that DocPrune improves throughput by 3.0x and 3.3x in the encoder and decoder, respectively, while boosting the F1 score by +1.0, achieving both higher accuracy and efficiency without any additional training.

Anticipatory Planning for Multimodal AI Agents

Recent advances in multimodal agents have improved computer-use interaction and tool-usage, yet most existing systems remain reactive, optimizing actions in isolation without reasoning about future states or long-term goals. This limits planning coherence and prevents agents from reliably solving high-level, multi-step tasks. We introduce TraceR1, a two-stage reinforcement learning framework that explicitly trains anticipatory reasoning by forecasting short-horizon trajectories before execution. The first stage performs trajectory-level reinforcement learning with rewards that enforce global consistency across predicted action sequences. The second stage applies grounded reinforcement fine-tuning, using execution feedback from frozen tool agents to refine step-level accuracy and executability. TraceR1 is evaluated across seven benchmarks, covering online computer-use, offline computer-use benchmarks, and multimodal tool-use reasoning tasks, where it achieves substantial improvements in planning stability, execution robustness, and generalization over reactive and single-stage baselines. These results show that anticipatory trajectory reasoning is a key principle for building multimodal agents that can reason, plan, and act effectively in complex real-world environments.

MiLDEdit: Reasoning-Based Multi-Layer Design Document Editing

Real-world design documents (e.g., posters) are inherently multi-layered, combining decoration, text, and images. Editing them from natural-language instructions requires fine-grained, layer-aware reasoning to identify relevant layers and coordinate modifications. Prior work largely overlooks multi-layer design document editing, focusing instead on single-layer image editing or multi-layer generation, which assume a flat canvas and lack the reasoning needed to determine what and where to modify. To address this gap, we introduce the Multi-Layer Document Editing Agent (MiLDEAgent), a reasoning-based framework that combines an RL-trained multimodal reasoner for layer-wise understanding with an image editor for targeted modifications. To systematically benchmark this setting, we introduce the MiLDEBench, a human-in-the-loop corpus of over 20K design documents paired with diverse editing instructions. The benchmark is complemented by a task-specific evaluation protocol, MiLDEEval, which spans four dimensions including instruction following, layout consistency, aesthetics, and text rendering. Extensive experiments on 14 open-source and 2 closed-source models reveal that existing approaches fail to generalize: open-source models often cannot complete multi-layer document editing tasks, while closed-source models suffer from format violations. In contrast, MiLDEAgent achieves strong layer-aware reasoning and precise editing, significantly outperforming all open-source baselines and attaining performance comparable to closed-source models, thereby establishing the first strong baseline for multi-layer document editing.

Text-Conditioned Background Generation for Editable Multi-Layer Documents

We present a framework for document-centric background generation with multi-page editing and thematic continuity. To ensure text regions remain readable, we employ a \emph{latent masking} formulation that softly attenuates updates in the diffusion space, inspired by smooth barrier functions in physics and numerical optimization. In addition, we introduce \emph{Automated Readability Optimization (ARO)}, which automatically places semi-transparent, rounded backing shapes behind text regions. ARO determines the minimal opacity needed to satisfy perceptual contrast standards (WCAG 2.2) relative to the underlying background, ensuring readability while maintaining aesthetic harmony without human intervention. Multi-page consistency is maintained through a summarization-and-instruction process, where each page is distilled into a compact representation that recursively guides subsequent generations. This design reflects how humans build continuity by retaining prior context, ensuring that visual motifs evolve coherently across an entire document. Our method further treats a document as a structured composition in which text, figures, and backgrounds are preserved or regenerated as separate layers, allowing targeted background editing without compromising readability. Finally, user-provided prompts allow stylistic adjustments in color and texture, balancing automated consistency with flexible customization. Our training-free framework produces visually coherent, text-preserving, and thematically aligned documents, bridging generative modeling with natural design workflows.

VisR-Bench: An Empirical Study on Visual Retrieval-Augmented Generation for Multilingual Long Document Understanding

Most organizational data in this world are stored as documents, and visual retrieval plays a crucial role in unlocking the collective intelligence from all these documents. However, existing benchmarks focus on English-only document retrieval or only consider multilingual question-answering on a single-page image. To bridge this gap, we introduce VisR-Bench, a multilingual benchmark designed for question-driven multimodal retrieval in long documents. Our benchmark comprises over 35K high-quality QA pairs across 1.2K documents, enabling fine-grained evaluation of multimodal retrieval. VisR-Bench spans sixteen languages with three question types (figures, text, and tables), offering diverse linguistic and question coverage. Unlike prior datasets, we include queries without explicit answers, preventing models from relying on superficial keyword matching. We evaluate various retrieval models, including text-based methods, multimodal encoders, and MLLMs, providing insights into their strengths and limitations. Our results show that while MLLMs significantly outperform text-based and multimodal encoder models, they still struggle with structured tables and low-resource languages, highlighting key challenges in multilingual visual retrieval.

AVA-Bench: Atomic Visual Ability Benchmark for Vision Foundation Models

The rise of vision foundation models (VFMs) calls for systematic evaluation. A common approach pairs VFMs with large language models (LLMs) as general-purpose heads, followed by evaluation on broad Visual Question Answering (VQA) benchmarks. However, this protocol has two key blind spots: (i) the instruction tuning data may not align with VQA test distributions, meaning a wrong prediction can stem from such data mismatch rather than a VFM' visual shortcomings; (ii) VQA benchmarks often require multiple visual abilities, making it hard to tell whether errors stem from lacking all required abilities or just a single critical one. To address these gaps, we introduce AVA-Bench, the first benchmark that explicitly disentangles 14 Atomic Visual Abilities (AVAs) -- foundational skills like localization, depth estimation, and spatial understanding that collectively support complex visual reasoning tasks. By decoupling AVAs and matching training and test distributions within each, AVA-Bench pinpoints exactly where a VFM excels or falters. Applying AVA-Bench to leading VFMs thus reveals distinctive "ability fingerprints," turning VFM selection from educated guesswork into principled engineering. Notably, we find that a 0.5B LLM yields similar VFM rankings as a 7B LLM while cutting GPU hours by 8x, enabling more efficient evaluation. By offering a comprehensive and transparent benchmark, we hope AVA-Bench lays the foundation for the next generation of VFMs.

CompBench: A Comparative Reasoning Benchmark for Multimodal LLMs

The ability to compare objects, scenes, or situations is crucial for effective decision-making and problem-solving in everyday life. For instance, comparing the freshness of apples enables better choices during grocery shopping, while comparing sofa designs helps optimize the aesthetics of our living space. Despite its significance, the comparative capability is largely unexplored in artificial general intelligence (AGI). In this paper, we introduce CompBench, a benchmark designed to evaluate the comparative reasoning capability of multimodal large language models (MLLMs). CompBench mines and pairs images through visually oriented questions covering eight dimensions of relative comparison: visual attribute, existence, state, emotion, temporality, spatiality, quantity, and quality. We curate a collection of around 40K image pairs using metadata from diverse vision datasets and CLIP similarity scores. These image pairs span a broad array of visual domains, including animals, fashion, sports, and both outdoor and indoor scenes. The questions are carefully crafted to discern relative characteristics between two images and are labeled by human annotators for accuracy and relevance. We use CompBench to evaluate recent MLLMs, including GPT-4V(ision), Gemini-Pro, and LLaVA-1.6. Our results reveal notable shortcomings in their comparative abilities. We believe CompBench not only sheds light on these limitations but also establishes a solid foundation for future enhancements in the comparative capability of MLLMs.

ARES: Alternating Reinforcement Learning and Supervised Fine-Tuning for Enhanced Multi-Modal Chain-of-Thought Reasoning Through Diverse AI Feedback

Large Multimodal Models (LMMs) excel at comprehending human instructions and demonstrate remarkable results across a broad spectrum of tasks. Reinforcement Learning from Human Feedback (RLHF) and AI Feedback (RLAIF) further refine LLMs by aligning them with specific preferences. These methods primarily use ranking-based feedback for entire generations. With advanced AI models (Teacher), such as GPT-4 and Claude 3 Opus, we can request various types of detailed feedback that are expensive for humans to provide. We propose a two-stage algorithm ARES that Alternates REinforcement Learning (RL) and Supervised Fine-Tuning (SFT). First, we request the Teacher to score how much each sentence contributes to solving the problem in a Chain-of-Thought (CoT). This sentence-level feedback allows us to consider individual valuable segments, providing more granular rewards for the RL procedure. Second, we ask the Teacher to correct the wrong reasoning after the RL stage. The RL procedure requires massive efforts for hyperparameter tuning and often generates errors like repetitive words and incomplete sentences. With the correction feedback, we stabilize the RL fine-tuned model through SFT. We conduct experiments on multi-model dataset ScienceQA and A-OKVQA to demonstrate the effectiveness of our proposal. ARES rationale reasoning achieves around 70% win rate against baseline models judged by GPT-4o. Additionally, we observe that the improved rationale reasoning leads to a 2.5% increase in inference answer accuracy on average for the multi-modal datasets.