Recent large language models (LLMs) with long Chain-of-Thought reasoning--such as DeepSeek-R1--have achieved impressive results on Olympiad-level mathematics benchmarks. However, they often rely on a narrow set of strategies and struggle with problems that require a novel way of thinking [33]. To systematically investigate these limitations, we introduce OMEGA--Out-of-distribution Math Problems Evaluation with 3 Generalization Axes--a controlled yet diverse benchmark designed to evaluate three axes of out-of-distribution generalization, inspired by Boden's typology of creativity [4]: (1) Exploratory--applying known problemsolving skills to more complex instances within the same problem domain; (2) Compositional--combining distinct reasoning skills, previously learned in isolation, to solve novel problems that require integrating these skills in new and coherent ways; and (3) Transformative--adopting novel, often unconventional strategies by moving beyond familiar approaches to solve problems more effectively. OMEGA consists of programmatically generated training-test pairs derived from templated problem generators across geometry, number theory, algebra, combinatorics, logic, and puzzles, with solutions verified using symbolic, numerical, or graphical methods. We evaluate frontier (or top-tier) LLMs and observe sharp performance degradation as problem complexity increases. Moreover, we fine-tune the Qwenseries models across all generalization settings and observe notable improvements in exploratory generalization, while compositional generalization remains limited and transformative reasoning shows little to no improvement. By isolating and quantifying these fine-grained failures, OMEGA lays the groundwork for advancing LLMs toward genuine mathematical creativity beyond mechanical proficiency.

Vision-Language Navigation (VLN) is a critical task for developing embodied agents that can follow natural language instructions to navigate in complex realworld environments. Recent advances driven by large pretrained models have significantly improved generalization and instruction grounding compared to traditional approaches. However, reasoning strategies in this task remain underexplored. Navigation is action-centric and long-horizon, while Chain-of-Thought (CoT) reasoning has mainly shown success in static tasks such as visual question answering. To address this gap, we conduct the first systematic evaluation of reasoning strategies, including No-Think (direct action prediction), Pre-Think (reasoning before action), and Post-Think (reasoning after action). Surprisingly, our findings reveal a Test-time Reasoning Collapse issue, where reasoning during testing degrades navigation accuracy, highlighting the challenges of integrating reasoning into embodied navigation.

Recent advancements in large language models have demonstrated how chain-of-thought (CoT) and reinforcement learning (RL) can improve performance. However, applying such reasoning strategies to the visual generation domain remains largely unexplored.

Vision-Language Navigation is a critical task for developing embodied agents that can follow natural language instructions to navigate in complex real-world environments. Recent advances by finetuning large pretrained models have significantly improved generalization and instruction grounding compared to traditional approaches. However, the role of reasoning strategies in navigation--an action-centric, long-horizon task--remains underexplored, despite Chain-of-Thought reasoning's demonstrated success in static tasks like question answering and visual reasoning. To address this gap, we conduct the first systematic evaluation of reasoning strategies for VLN, including No-Think (direct action prediction), Pre-Think (reason before action), and Post-Think (reason after action). Surprisingly, our findings reveal the Inference-time Reasoning Collaps issue, where inference-time reasoning degrades navigation accuracy, highlighting the challenges of integrating reasoning into VLN. Based on this insight, we propose Aux-Think, a framework that trains models to internalize structured reasoning patterns through CoT supervision during training, while preserving No-Think inference for efficient action prediction. To support this framework, we release R2R-CoT-320k, a large-scale Chain-of-Thought annotated dataset. Empirically, Aux-Think significantly reduces training effort without compromising performance.

Standard chain-of-thought reasoning generates a solution in a single forward pass, committing irrevocably to each token and lacking a mechanism to recover from early errors. We introduce Inference-Time Rethinking, a generative framework that enables iterative self-correction by decoupling declarative latent thought vectors from procedural generation. We factorize reasoning into a continuous latent thought vector (what to reason about) and a decoder that verbalizes the trace conditioned on this vector (how to reason). Beyond serving as a declarative buffer, latent thought vectors compress the reasoning structure into a continuous representation that abstracts away surface-level token variability, making gradient-based optimization over reasoning strategies well-posed. Our prior model maps unstructured noise to a learned manifold of valid reasoning patterns, and at test time we employ a Gibbs-style procedure that alternates between generating a candidate trace and optimizing the latent vector to better explain that trace, effectively navigating the latent manifold to refine the reasoning strategy. Training a 0.2B-parameter model from scratch on GSM8K, our method with 30 rethinking iterations surpasses baselines with 10 to 15 times more parameters, including a 3B counterpart. This result demonstrates that effective mathematical reasoning can emerge from sophisticated inference-time computation rather than solely from massive parameter counts.

We are interested in understanding how well Transformer language models (TLMs) can perform reasoning tasks when trained on knowledge encoded in the form of natural language. We investigate their systematic generalization abilities on a logical reasoning task in natural language, which involves reasoning over relationships between entities grounded in first-order logical proofs. Specifically, we perform soft theorem-proving by leveraging TLMs to generate natural language proofs. We test the generated proofs for logical consistency, along with the accuracy of the final inference. We observe length-generalization issues when evaluated on longer-than-trained sequences. However, we observe TLMs improve their generalization performance after being exposed to longer, exhaustive proofs.

Large language models (LLMs) are post-trained through reinforcement learning (RL) to evolve into Reasoning Language Models (RLMs), where the hallmark of this advanced reasoning is ``aha'' moments when they start to perform strategies, such as self-reflection and deep thinking, within chain of thoughts (CoTs). Motivated by this, this paper proposes a novel reinforced strategy injection mechanism (rSIM), that enables any LLM to become an RLM by employing a small planner to guide the LLM's CoT through the adaptive injection of reasoning strategies. To achieve this, the planner (leader agent) is jointly trained with an LLM (follower agent) using multi-agent RL (MARL), based on a leader-follower framework and straightforward rule-based rewards. Experimental results show that rSIM enables Qwen2.5-0.5B to become an RLM and significantly outperform Qwen2.5-14B. Moreover, the planner is generalizable: it only needs to be trained once and can be applied as a plug-in to substantially improve the reasoning capabilities of existing LLMs. In addition, the planner supports continual learning across various tasks, allowing its planning abilities to gradually improve and generalize to a wider range of problems.

While the reasoning capabilities of Large Language Models (LLMs) excel in analytical tasks such as mathematics and code generation, their utility for abstractive summarization remains widely assumed but largely unverified. To bridge this gap, we first tailor general reasoning strategies to the summarization domain. We then conduct a systematic, large scale comparative study of 8 reasoning strategies and 3 Large Reasoning Models (LRMs) across 8 diverse datasets, assessing both summary quality and faithfulness. Our findings show that reasoning is not a universal solution and its effectiveness is highly dependent on the specific strategy and context. Specifically, we observe a trade-off between summary quality and factual faithfulness: explicit reasoning strategies tend to improve fluency at the expense of factual grounding, while implicit reasoning in LRMs exhibits the inverse pattern. Furthermore, increasing an LRM's internal reasoning budget does not improve, and can even hurt, factual consistency, suggesting that effective summarization demands faithful compression rather than creative over-thinking.

Large language models (LLMs) are increasingly deployed in settings where reasoning, such as multi-step problem solving and chain-of-thought, is essential. Yet, current evaluation practices overwhelmingly report single-run accuracy while ignoring the intrinsic uncertainty that naturally arises from stochastic decoding. This omission creates a blind spot because practitioners cannot reliably assess whether a method's reported performance is stable, reproducible, or cost-consistent. We introduce ReasonBENCH, the first benchmark designed to quantify the underlying instability in LLM reasoning. ReasonBENCH provides (i) a modular evaluation library that standardizes reasoning frameworks, models, and tasks, (ii) a multi-run protocol that reports statistically reliable metrics for both quality and cost, and (iii) a public leaderboard to encourage variance-aware reporting. Across tasks from different domains, we find that the vast majority of reasoning strategies and models exhibit high instability. Notably, even strategies with similar average performance can display confidence intervals up to four times wider, and the top-performing methods often incur higher and less stable costs. Such instability compromises reproducibility across runs and, consequently, the reliability of reported performance. To better understand these dynamics, we further analyze the impact of prompts, model families, and scale on the trade-off between solve rate and stability. Our results highlight reproducibility as a critical dimension for reliable LLM reasoning and provide a foundation for future reasoning methods and uncertainty quantification techniques. ReasonBENCH is publicly available at https://github.com/au-clan/ReasonBench .

Large language models (LLMs) have achieved impressive results in code generation, yet struggle with program verification, especially in interactive proof frameworks such as Lean4. A central challenge is scalability: verified synthesis requires not just code, but also precise specifications and correctness proofs, and existing approaches rarely span all three domains. We present BRIDGE, the first systematic study of structured prompting for scalable verified program generation. BRIDGE decomposes verification into three interconnected domains: Code (executable implementations), Specifications (formal intent statements), and Proofs (constructive correctness arguments). Our key idea is to elicit distinct reasoning behaviors functional, specification-driven, and proof-oriented as intermediate representations that preserve semantic structure and connect these domains. Through systematic ablations, we show that this approach substantially improves both accuracy and efficiency beyond standard error feedback methods. For example, functional reasoning improves correctness of code in formal languages (Lean4) by nearly 1.5x (pass@5) over direct baselines. In inference-time compute, functional reasoning is also 2x more efficient, achieving higher pass rates with fewer generations and lower total sampling budgets. Similarly, we find that specification-driven prompting boosts Python coding pass rates by up to 17.5%. These findings suggest that structured domain alignment is a promising direction for advancing verified synthesis. BRIDGE establishes a foundation for training via expert iteration or RLVR, enabling models to internalize these reasoning strategies across code, specifications, and proofs.