Large Language Models (LLMs) have shown strong reasoning capabilities, particularly when enhanced through Reinforcement Learning (RL). While prior work has successfully applied RL to mathematical reasoning -- where rules and correctness are well-defined -- generalizing these methods to broader reasoning domains remains challenging due to limited data, the lack of verifiable reward structures, and diverse task requirements. In this work, we propose NEMOTRON-CROSSTHINK, a framework that systematically incorporates multi-domain corpora, including both synthetic and real-world question-answer pairs, into RL training to improve generalization across diverse reasoning tasks. NEMOTRON-CROSSTHINK addresses key challenges by (1) incorporating data from varied sources spanning STEM, humanities, social sciences, etc.; (2) applying structured templates (e.g., multiple-choice and open-ended) to control answer-space complexity; (3) filtering for verifiable answers; and (4) optimizing data blending strategies that utilizes data from multiple sources effectively. Our approach enables scalable and verifiable reward modeling beyond mathematics and demonstrates improved accuracies on both math (MATH-500: +30.1%, AMC23:+27.5%) and non-math reasoning benchmarks (MMLU-PRO: +12.8%, GPQA-DIAMOND: +11.3%, AGIEVAL: +15.1%, SUPERGPQA: +3.8%). Moreover, NEMOTRON-CROSSTHINK exhibits significantly improved response efficiency -- using 28% fewer tokens for correct answers -- highlighting more focused and effective reasoning. Through NEMOTRON-CROSSTHINK, we demonstrate that integrating multi-domain, multi-format data in RL leads to more accurate, efficient, and generalizable LLMs.

Geometry problem-solving (GPS), a challenging task requiring both visual comprehension and symbolic reasoning, effectively measures the reasoning capabilities of multimodal large language models (MLLMs). Humans exhibit strong reasoning ability in this task through accurate identification and adaptive application of geometric principles within visual contexts. However, existing benchmarks fail to jointly assess both dimensions of the human-like geometric reasoning mechanism in MLLMs, remaining a critical gap in assessing their ability to tackle GPS. To this end, we introduce GeoSense, the first comprehensive bilingual benchmark designed to systematically evaluate the geometric reasoning abilities of MLLMs through the lens of geometric principles. GeoSense features a five-level hierarchical framework of geometric principles spanning plane and solid geometry, an intricately annotated dataset of 1,789 problems, and an innovative evaluation strategy. Through extensive experiments on GeoSense with various open-source and closed-source MLLMs, we observe that Gemini-2.0-pro-flash performs best, achieving an overall score of $65.3$. Our in-depth analysis reveals that the identification and application of geometric principles remain a bottleneck for leading MLLMs, jointly hindering their reasoning abilities. These findings underscore GeoSense's potential to guide future advancements in MLLMs' geometric reasoning capabilities, paving the way for more robust and human-like reasoning in artificial intelligence.

This paper presents our winning submission to the AI Mathematical Olympiad - Progress Prize 2 (AIMO-2) competition. Our recipe for building state-of-the-art mathematical reasoning models relies on three key pillars. First, we create a large-scale dataset comprising 540K unique high-quality math problems, including olympiad-level problems, and their 3.2M long-reasoning solutions. Second, we develop a novel method to integrate code execution with long reasoning models through iterative training, generation, and quality filtering, resulting in 1.7M high-quality Tool-Integrated Reasoning solutions. Third, we create a pipeline to train models to select the most promising solution from many candidates. We show that such generative solution selection (GenSelect) can significantly improve upon majority voting baseline. Combining these ideas, we train a series of models that achieve state-of-the-art results on mathematical reasoning benchmarks. To facilitate further research, we release our code, models, and the complete OpenMathReasoning dataset under a commercially permissive license.

Autonomous AI systems reveal foundational limitations in deterministic, human-authored computing architectures. This paper presents Cognitive Silicon: a hypothetical full-stack architectural framework projected toward 2035, exploring a possible trajectory for cognitive computing system design. The proposed architecture would integrate symbolic scaffolding, governed memory, runtime moral coherence, and alignment-aware execution across silicon-to-semantics layers. Our design grammar has emerged from dialectical co-design with LLMs under asymmetric epistemic conditions--creating structured friction to expose blind spots and trade-offs. The envisioned framework would establish mortality as a natural consequence of physical constraints, non-copyable tacit knowledge, and non-cloneable identity keys as cognitive-embodiment primitives. Core tensions (trust/agency, scaffolding/emergence, execution/governance) would function as central architectural pressures rather than edge cases. The architecture theoretically converges with the Free Energy Principle, potentially offering a formal account of how cognitive systems could maintain identity through prediction error minimization across physical and computational boundaries. The resulting framework aims to deliver a morally tractable cognitive infrastructure that could maintain human-alignment through irreversible hardware constraints and identity-bound epistemic mechanisms resistant to replication or subversion.

While recent advancements in robotic manipulation video synthesis have shown promise, significant challenges persist in ensuring effective instruction-following and achieving high visual quality. Recent methods, like RoboDreamer, utilize linguistic decomposition to divide instructions into separate lower-level primitives, conditioning the world model on these primitives to achieve compositional instruction-following. However, these separate primitives do not consider the relationships that exist between them. Furthermore, recent methods neglect valuable visual guidance, including depth and semantic guidance, both crucial for enhancing visual quality. This paper introduces ManipDreamer, an advanced world model based on the action tree and visual guidance. To better learn the relationships between instruction primitives, we represent the instruction as the action tree and assign embeddings to tree nodes, each instruction can acquire its embeddings by navigating through the action tree. The instruction embeddings can be used to guide the world model. To enhance visual quality, we combine depth and semantic guidance by introducing a visual guidance adapter compatible with the world model. This visual adapter enhances both the temporal and physical consistency of video generation. Based on the action tree and visual guidance, ManipDreamer significantly boosts the instruction-following ability and visual quality. Comprehensive evaluations on robotic manipulation benchmarks reveal that ManipDreamer achieves large improvements in video quality metrics in both seen and unseen tasks, with PSNR improved from 19.55 to 21.05, SSIM improved from 0.7474 to 0.7982 and reduced Flow Error from 3.506 to 3.201 in unseen tasks, compared to the recent RoboDreamer model. Additionally, our method increases the success rate of robotic manipulation tasks by 2.5% in 6 RLbench tasks on average.

Reasoning in large language models (LLMs) tends to produce substantially longer token generation sequences than simpler language modeling tasks. This extended generation length reflects the multi-step, compositional nature of reasoning and is often correlated with higher solution accuracy. From an efficiency perspective, longer token generation exacerbates the inherently sequential and memory-bound decoding phase of LLMs. However, not all parts of this expensive reasoning process are equally difficult to generate. We leverage this observation by offloading only the most challenging parts of the reasoning process to a larger, more capable model, while performing most of the generation with a smaller, more efficient model; furthermore, we teach the smaller model to identify these difficult segments and independently trigger offloading when needed. To enable this behavior, we annotate difficult segments across 18k reasoning traces from the OpenR1-Math-220k chain-of-thought (CoT) dataset. We then apply supervised fine-tuning (SFT) and reinforcement learning fine-tuning (RLFT) to a 1.5B-parameter reasoning model, training it to learn to offload the most challenging parts of its own reasoning process to a larger model. This approach improves AIME24 reasoning accuracy by 24% and 28.3% while offloading 1.35% and 5% of the generated tokens respectively. We open-source our SplitReason model, data, code and logs.

Automatic extraction of definitions from legal texts is critical for enhancing the comprehension and clarity of complex legal corpora such as the United States Code (U.S.C.). We present an advanced NLP system leveraging transformer-based architectures to automatically extract defined terms, their definitions, and their scope from the U.S.C. We address the challenges of automatically identifying legal definitions, extracting defined terms, and determining their scope within this complex corpus of over 200,000 pages of federal statutory law. Building upon previous feature-based machine learning methods, our updated model employs domain-specific transformers (Legal-BERT) fine-tuned specifically for statutory texts, significantly improving extraction accuracy. Our work implements a multi-stage pipeline that combines document structure analysis with state-of-the-art language models to process legal text from the XML version of the U.S. Code. Each paragraph is first classified using a fine-tuned legal domain BERT model to determine if it contains a definition. Our system then aggregates related paragraphs into coherent definitional units and applies a combination of attention mechanisms and rule-based patterns to extract defined terms and their jurisdictional scope. The definition extraction system is evaluated on multiple titles of the U.S. Code containing thousands of definitions, demonstrating significant improvements over previous approaches. Our best model achieves 96.8% precision and 98.9% recall (98.2% F1-score), substantially outperforming traditional machine learning classifiers. This work contributes to improving accessibility and understanding of legal information while establishing a foundation for downstream legal reasoning tasks.

--Vision systems are increasingly deployed in critical domains such as surveillance, law enforcement, and transportation. However, their vulnerabilities to rare or unforeseen scenarios pose significant safety risks. T o address these challenges, we introduce Context-A wareness and Interpretability of Rare Occurrences (CAIRO), an ontology-based human-assistive discovery framework for failure cases (or CP - Critical Phenomena) detection and formalization. CAIRO by design incentivizes human-in-the-loop for testing and evaluation of criticality that arises from misdetections, adversarial attacks, and hallucinations in AI black-box models. Our robust analysis of object detection model(s) failures in automated driving systems (ADS) showcases scalable and interpretable ways of formalizing the observed gaps between camera perception and real-world contexts, resulting in test cases stored as explicit knowledge graphs (in OWL/XML format) amenable for sharing, downstream analysis, logical reasoning, and accountability. I NTRODUCTION Formal verification techniques are a norm in chip design, but they remain elusive in computer vision (CV) applications. The reason being CV applications are deemed open-ended, often trained on millions of data and billions of parameters to learn a few hundreds of labels. Finetuning practices are commonly used to tailor them to specific needs, but with no standard testing procedures in place providing guidance for their application to ensure fail-safe behaviors, critical systems like Autonomous V ehicles (A V) are bound to fail [1].

Predominant LLMs focus on high-resource languages while leaving low-resource languages, particularly those in Southeast Asia (SEA), underrepresented. In addition, those models are general-purpose and pay limited attention to the e-commerce domain. To overcome these limitations, we introduce Compass-v2, a lightweight Mixture-of-Experts (MoE) model specifically designed for Southeast Asian languages and e-commerce applications. To balance model performance and inference cost, the model is designed with 30B total parameters and 5B active parameters, incorporating both fine-grained and shared expert modules. To enhance multilingual performance, we curated and constructed a high-quality, industry-leading SEA dataset, to the best of our knowledge. To boost performance in the e-commerce domain, we built a dataset comprising hundreds of billions of tokens, sourced through external data mining and internal platform collection. Besides, we pioneered a hybrid reasoning model that supports both fast thinking and deep thinking within a unified framework to enhance the reasoning capabilities, diverging from the conventional industry practice of deploying two separate models. Through extensive experimental evaluations, our model demonstrates state-of-the-art SEA multilingual and e-commerce performance among sub-30B models, while maintaining significantly lower inference cost.

Large Language Models (LLMs) have made remarkable breakthroughs in reasoning, yet continue to struggle with hallucinations, logical errors, and inability to self-correct during complex multi-step tasks. Current approaches like chain-of-thought prompting offer limited reasoning capabilities that fail when precise step validation is required. We propose Environment Augmented Generation (EAG), a framework that enhances LLM reasoning through: (1) real-time environmental feedback validating each reasoning step, (2) dynamic branch exploration for investigating alternative solution paths when faced with errors, and (3) experience-based learning from successful reasoning trajectories. Unlike existing methods, EAG enables deliberate backtracking and strategic replanning through tight integration of execution feedback with branching exploration. Our a1-32B model achieves state-of-the-art performance among similar-sized models across all benchmarks, matching larger models like o1 on competition mathematics while outperforming comparable models by up to 24.4 percentage points. Analysis reveals EAG's distinctive scaling pattern: initial token investment in environment interaction yields substantial long-term performance dividends, with advantages amplifying proportionally to task complexity. EAG's theoretical framework demonstrates how environment interactivity and systematic branch exploration together establish a new paradigm for reliable machine reasoning, particularly for problems requiring precise multi-step calculation and logical verification.