Test-time scaling (TTS) has been shown to improve the performance of large language models (LLMs) by sampling and aggregating diverse reasoning paths. However, existing research has overlooked a critical issue: selection bias of reasoning strategies during scaling. Specifically, when generating reasoning processes, LLMs tend to follow certain strategies (e.g., algebraic solutions for math problems) while neglecting other valid alternatives (e.g., geometric solutions), resulting in insufficient exploration of the solution space. To further understand the impact of this bias, we present a theoretical analysis that reveals when it undermines the effectiveness of test-time scaling. Motivated by this theoretical insight, we introduce TTS-Uniform, a framework designed to mitigate the selection bias of reasoning strategies. It (i) identifies potential strategies, (ii) uniformly allocates the sampling budget across them, and (iii) filters out unstable strategies prior to aggregation. Experimental results show that TTS-Uniform significantly enhances scaling effectiveness across multiple mainstream LLMs and benchmark datasets. Code is available at https://github.com/zongqianwu/Uniform-TTS. Chain-of-thought (CoT) (Wei et al., 2022; Kojima et al., 2022) enhances the reasoning capabilities of large language models (LLMs) by explicitly unfolding intermediate steps (i.e., reasoning paths) before arriving at the final answer. Building on CoT, test-time scaling (TTS) (Zhang et al., 2025; Ji et al., 2025) further improves performance by sampling and aggregating diverse paths. However, existing TTS research (Wang et al., 2022; Snell et al., 2024) overlooks a critical limitation of CoT, which in turn constrains the effectiveness of scaling. Specifically, when tackling a problem, CoT reasoning tends to follow certain strategies while neglecting other valid alternatives.

The remarkable reasoning capability of large language models (LLMs) stems from cognitive behaviors that emerge through reinforcement with verifiable rewards. This work investigates how to transfer this principle to Multimodal LLMs (MLLMs) to unlock advanced visual reasoning. We introduce a two-stage paradigm built on Qwen2.5-VL-7B: a massive linguistic cold-start fine-tuning, followed by multimodal reinforcement learning (RL) spanning nearly 1,000 steps, surpassing all previous open-source efforts in scale. This pioneering work reveals three fundamental insights: 1) Behavior transfer emerges surprisingly early in cold start due to linguistic mental imagery. 2) Cold start broadly memorizes visual behaviors, while RL critically discerns and scales up effective patterns. 3) Transfer strategically favors high-utility behaviors such as visual reflection. Our resulting model, Open-Vision-Reasoner (OVR), achieves state-of-the-art performance on a suite of reasoning benchmarks, including 95.3% on MATH500, 51.8% on MathVision and 54.6% on MathVerse. We release our model, data, and training dynamics to catalyze the development of more capable, behavior-aligned multimodal reasoners.

Reward models are essential for aligning language model outputs with human preferences, yet existing approaches often lack both controllability and interpretability. These models are typically optimized for narrow objectives, limiting their generalizability to broader downstream tasks. Moreover, their scalar outputs are difficult to interpret without contextual reasoning. To address these limitations, we introduce $\shortmethodname$, a novel reward modeling framework that is rubric-agnostic, generalizable across evaluation dimensions, and provides interpretable, reasoned score assignments. $\shortmethodname$ enables more transparent and flexible evaluation of language models, supporting robust alignment with diverse human values and use cases. Our models, data, and code are available as open source at https://github.com/rubricreward/r3.

Reinforcement learning (RL) has become a standard paradigm for refining large language models (LLMs) beyond pre-training and instruction tuning. A prominent line of work is RL with verifiable rewards (RLVR), which leverages automatically verifiable outcomes (e.g., correctness or executability) to generate reward signals. While efficient, this framework faces two key limitations: First, its binary feedback is too sparse to capture the quality of the reasoning process. Second, its coarse-grained rewards potentially lead to vanishing gradients. Inspired by observations from human learning, we introduce a RL technique that integrates verifiable outcomes with the model's own confidence estimates. This joint design enriches the reward signal, providing finer-grained feedback and implicitly supervising the reasoning process. Experimental results demonstrate that our proposed method enhances RL performance across multiple datasets and reduces token consumption during inference, while incurring negligible additional training cost. Moreover, it can be used as a plug-in module to enhance other state-of-the-art RL methods.

Model editing, the process of efficiently modifying factual knowledge in pre-trained language models, is critical for maintaining their accuracy and relevance. However, existing editing methods often introduce unintended side effects, degrading model performance in unpredictable ways. While much research has focused on improving editing algorithms, the role of the target knowledge's intrinsic properties remains a significant, underexplored factor. This paper addresses this gap by first proposing the ``Knowledge Spectrum,'' a systematic framework for categorizing knowledge based on its real-world popularity, the model's pre-edit familiarity, and the linguistic structure of the eliciting question. Our empirical analysis reveals that these characteristics are strong predictors of editing success and stability. Informed by these findings, we introduce the ``Knowledge-Diagnostic Framework,'' an adaptive strategy that tailors editing intensity to the diagnosed difficulty of a knowledge item. We demonstrate that this framework significantly improves success rates for challenging edits while optimizing computational resources. Our work provides a more comprehensive understanding of the factors governing model editing.

LLMs suffer from critical reasoning issues such as unfaithfulness, bias, and inconsistency, since they lack robust causal underpinnings and may rely on superficial correlations rather than genuine understanding. Successive LRMs have emerged as a promising alternative, leveraging advanced training techniques such as reinforcement learning (RL) and distillation to improve task accuracy. However, the impact of these training methods on causality remains largely unexplored. In this study, we conduct a systematic causal analysis on LLMs and LRMs, examining structural causal models (SCMs) of four key variables: problem instruction (Z), thinking process (T), reasoning steps (X), and answer (Y). Our findings reveal that RLVR-trained LRMs exhibit enhanced causal reasoning capabilities, aligning more closely with ideal causal structures, while LLMs and distilled LRMs fail to address causality-related deficiencies. Our further investigation indicates that RLVR reduces spurious correlations and strengthens genuine causal patterns, thereby mitigating unfaithfulness and bias. In addition, our inspection on the dynamics of the RLVR training process observes a high correlation between reduced spurious features and improved causal structures, where the causal relationships consistently improve in the training process. This study contributes to the understanding of causality in reasoning models, highlights the critical role of RLVR in enhancing causal reasoning, and provides insights for designing future AI systems with stronger causal foundations. We release our code and data at https://github.com/Harryking1999/CoT_Causal_Analysis.

Relational object rearrangement (ROR) tasks (e.g., insert flower to vase) require a robot to manipulate objects with precise semantic and geometric reasoning. Existing approaches either rely on pre-collected demonstrations that struggle to capture complex geometric constraints or generate goal-state observations to capture semantic and geometric knowledge, but fail to explicitly couple object transformation with action prediction, resulting in errors due to generative noise. To address these limitations, we propose Imagine2Act, a 3D imitation-learning framework that incorporates semantic and geometric constraints of objects into policy learning to tackle high-precision manipulation tasks. We first generate imagined goal images conditioned on language instructions and reconstruct corresponding 3D point clouds to provide robust semantic and geometric priors. These imagined goal point clouds serve as additional inputs to the policy model, while an object-action consistency strategy with soft pose supervision explicitly aligns predicted end-effector motion with generated object transformation. This design enables Imagine2Act to reason about semantic and geometric relationships between objects and predict accurate actions across diverse tasks. Experiments in both simulation and the real world demonstrate that Imagine2Act outperforms previous state-of-the-art policies. More visualizations can be found at https://sites.google.com/view/imagine2act.

Embodied agents operating in the physical world must make decisions that are not only effective but also safe, spatially coherent, and grounded in context. While recent advances in large multimodal models (LMMs) have shown promising capabilities in visual understanding and language generation, their ability to perform structured reasoning for real-world embodied tasks remains underexplored. In this work, we aim to understand how well foundation models can perform step-by-step reasoning in embodied environments. To this end, we propose the Foundation Model Embodied Reasoning (FoMER) benchmark, designed to evaluate the reasoning capabilities of LMMs in complex embodied decision-making scenarios. Our benchmark spans a diverse set of tasks that require agents to interpret multimodal observations, reason about physical constraints and safety, and generate valid next actions in natural language. We present (i) a large-scale, curated suite of embodied reasoning tasks, (ii) a novel evaluation framework that disentangles perceptual grounding from action reasoning, and (iii) empirical analysis of several leading LMMs under this setting. Our benchmark includes over 1.1k samples with detailed step-by-step reasoning across 10 tasks and 8 embodiments, covering three different robot types. Our results highlight both the potential and current limitations of LMMs in embodied reasoning, pointing towards key challenges and opportunities for future research in robot intelligence. Our data and code will be made publicly available.

The debate around Artificial General Intelligence (AGI) remains open due to two fundamentally different goals: replicating human-like performance versus replicating human-like cognitive processes. We argue that current performance-based definitions are inadequate because they provide no clear, mechanism-focused roadmap for research, and they fail to properly define the qualitative nature of genuine intelligence. Drawing inspiration from the human brain, we propose a new paradigm that shifts the focus from external mimicry to the development of foundational cognitive architectures. We define True Intelligence (TI) as a system characterized by six core components: embodied sensory fusion, core directives, dynamic schemata creation, a highly-interconnected multi-expert architecture, an orchestration layer, and lastly, the unmeasurable quality of Interconnectedness, which we hypothesize results in consciousness and a subjective experience. We propose a practical, five-level taxonomy of AGI based on the number of the first five measurable components a system exhibits. This framework provides a clear path forward with developmental milestones that directly address the challenge of building genuinely intelligent systems. We contend that once a system achieves Level-5 AGI by implementing all five measurable components, the difference between it and TI remains as a purely philosophical debate. For practical purposes - and given theories indicate consciousness is an emergent byproduct of integrated, higher-order cognition - we conclude that a fifth-level AGI is functionally and practically equivalent to TI. This work synthesizes diverse insights from analytical psychology, schema theory, metacognition, modern brain architectures and latest works in AI to provide the first holistic, mechanism-based definition of AGI that offers a clear and actionable path for the research community.

Learning visuomotor policies via imitation has proven effective across a wide range of robotic domains. However, the performance of these policies is heavily dependent on the number of training demonstrations, which requires expensive data collection in the real world. In this work, we aim to reduce data collection efforts when learning visuomotor robot policies by leveraging existing or cost-effective data from a wide range of embodiments, such as public robot datasets and the datasets of humans playing with objects (human data from play). Our approach leverages two key insights. First, we use optic flow as an embodiment-agnostic action representation to train a World Model (WM) across multi-embodiment datasets, and finetune it on a small amount of robot data from the target embodiment. Second, we develop a method, Latent Policy Steering (LPS), to improve the output of a behavior-cloned policy by searching in the latent space of the WM for better action sequences. In real world experiments, we observe significant improvements in the performance of policies trained with a small amount of data (over 50% relative improvement with 30 demonstrations and over 20% relative improvement with 50 demonstrations) by combining the policy with a WM pretrained on two thousand episodes sampled from the existing Open X-embodiment dataset across different robots or a cost-effective human dataset from play.