Long-horizon LLM agents execute sequences of reasoning, tool-use, and environment-interaction steps, making end-to-end latency a central deployment bottleneck. Speculative agents reduce this latency by leveraging a faster draft agent to propose the action trajectory while a stronger target agent verifies each action to ensure lossless action trajectories. Recent systems adapt how many future actions a preselected draft agent proposes before verification, but leave the drafter itself unchanged across tasks and episodes. Our experiments show that the best drafter varies across tasks and trajectory prefixes, revealing substantial latency headroom from adaptive drafter selection. This motivates the question: which draft agent should speculate for a given task and target agent to achieve low end-to-end latency? We propose AdaSpec, an adaptive routing framework for speculative agents with an objective distinct from standard model routing: selecting the drafter that minimizes realized speculative-execution latency for a target agent, rather than selecting the model that produces the final answer. AdaSpec profiles a pool of candidate draft-agent configurations for each target agent and trains a lightweight, latency-supervised router to rank them by measured speculative latency. The router supports both query-level routing and episode-level rerouting before each draft-and-check segment, while the target agent remains responsible for verifying actions. AdaSpec improves fixed-drafter baselines across multiple datasets and target-agent settings.

Multimodal Large Language Models (MLLMs) show strong potential for cross-modal understanding by integrating powerful language models with multimodal encoders. However, extending MLLMs to handle a diverse range of modalities introduces two critical and intertwined challenges: (1) the reliance on fully paired multimodal data, often scarce or costly to acquire across all modalities, and (2) the computational inefficiency from processing numerous modality tokens and requiring substantial model updates for each new modality. To address these challenges, we enable MLLMs to handle missing modalities by generating representations for absent inputs. Furthermore, recognizing that an increasing number of modalities leads to linearly scaling token counts and that lengthy generated sequences can hinder performance, we employ a dual-stage compression mechanism. It first reduces the number of tokens per modality and then condenses information from multiple modalities into a single, compact token sequence. This culminates in Flex-M³, a novel MLLM framework designed for flexible and efficient learning across arbitrary combinations of modalities. Experiments across diverse multimodal benchmarks and backbones demonstrate that Flex-M³ robustly handles varied modality inputs and scales efficiently. Notably, Flex-M outperforms its counterpart trained on only full-modality data, with consistent improvements of 2.29%, 3.15%, 11.01% on multimodal reasoning tasks NExT-QA, MUSIC-AVQA, SQA3D. Moreover, Flex-M³ demonstrates superior robustness during inference, even when a high proportion of modalities are missing from the input samples, showcasing its capacity for complex, data-scarce multimodal applications.

End-to-end GUI agents for real desktop environments require large amounts of high-quality interaction data, yet collecting human demonstrations is expensive and existing synthetic pipelines often suffer from limited task diversity or noisy, goal-drifting trajectories. We present a trajectory expansion framework Anchor that bootstraps scalable desktop supervision from a small set of verified seed demonstrations. Starting from each seed, we identify branch points that correspond to meaningful state changes and propose new, state-grounded task variants conditioned on the current GUI context. An executing agent then follows the proposed instructions to generate new trajectories, while a verifier enforces task completion via state-aware checks and trajectory-level consistency. To improve supervision quality, we further apply task-conditioned step-level filtering to remove ungrounded actions and denoise post-branch segments to maintain coherent intent. Experiments on standard desktop benchmarks, OSWorld and WindowsAgentArena, show that models fine-tuned on our expanded corpus achieve consistent improvements over zero-shot agents and representative synthesis baselines, and generalize across applications and operating systems.

Large Language Models (LLMs), when used in educational settings without pedagogical fine-tuning, often provide immediate answers rather than guiding students through the problem-solving process. This approach falls short of pedagogically best practices and limits their effectiveness as educational tools. We term the objective of training LLMs to emulate effective teaching strategies as ‘pedagogical alignment.’ In this paper, we investigate Learning from Human Preferences () algorithms to achieve this alignment objective. A key challenge in this process is the scarcity of high-quality preference datasets to guide the alignment. To address this, we propose a novel approach for constructing a large-scale dataset using synthetic data generation techniques, eliminating the need for time-consuming and costly manual annotation. Leveraging this dataset, our experiments with Llama and Mistral models demonstrate that LHP methods outperform standard supervised fine-tuning (SFT), improving pedagogical alignment accuracy by 13.1% and 8.7% respectively.Existing evaluation methods also lack quantitative metrics to adequately measure the pedagogical alignment of LLMs. To address this gap, we propose novel perplexity-based metrics that quantify LLMs’ tendency to provide scaffolded guidance versus direct answers, offering a robust measure of pedagogical alignment. Our analysis provides compelling evidence for the superiority of methods over SFT in optimizing LLMs’ behavior, underscoring the potential of methods in better aligning LLMs with educational objectives and fostering effective learning experiences. Code and models are available here.