Deep Research Agents (DR-Agents) extend LLMs with multi-step retrieval, tool use, and intermediate computation to solve complex problems. However, their execution logs often contain redundant context, noisy evidence, and brittle dependencies among intermediate decisions. As a result, final answers can be sensitive to irrelevant or weakly verified information. We propose DualResearch, a log-based post-processing framework for answer reconstruction. Given a query and a completed DR-Agent log, DualResearch normalizes the log into event records and constructs two evidence graphs: a breadth semantic graph for declarative evidence and a depth causal graph for procedural provenance. It then retrieves breadth nodes and depth paths, assigns higher weights to retrieval channels with more concentrated retained candidate distributions, and jointly considers relevance and cross-view agreement to select evidence for reconstruction; the concentration measure reflects retrieval confidence rather than factual correctness. Experiments on HLE, GAIA, and GPQA show consistent accuracy improvements over underlying DR-Agents and controlled log-reuse baselines.

Reward modeling aligns language models with human preferences, yet coding tasks, requiring strict logic and syntax, pose challenges due to their domain diversity. In this paper, we propose RewardCode, a generalist code reward model trained via a two-stage pipeline, enabling scalable and precise evaluation through principle-based scoring and unit test generation. Specifically, the first stage combines Structured Fine-Tuning and Rejective Fine-Tuning to distill cross-domain summarization and code evaluation capabilities from advanced LLMs. The second stage introduces Pairwise-GRPO, a reinforcement learning method that uses preference pairs to improve solution discrimination, while ensuring the generation of consistent and verifiable unit tests. Further, to train a cross-domain code reward model, we construct CodePair-18K, a dataset of verifiable code preference pairs and executable unit tests. Experiments on multiple benchmarks show that RewardCode outperforms strong baselines in evaluation accuracy, verifying its effectiveness in advancing general-purpose code LLMs.

Frontier scientific reasoning remains a major challenge for large language models (LLMs), where even the strongest commercial systems fall short of expert-level performance. A closer look at model behavior reveals substantial complementarity that single-model evaluation hides: different frontier models excel on different question types, and no single model captures the full picture. We present SciOrch, a framework that trains a lightweight 8B model to orchestrate frontier LLMs for scientific reasoning. The orchestrator decomposes each question, delegates sub-problems to selected commercial models through API calls, and synthesizes a final answer. Training such an orchestrator is fundamentally harder than conventional agentic RL: each action triggers an API call that is expensive in both dollar cost and latency, making standard online rollouts infeasible. We address this with MCTS-based approach, producing diverse orchestration trajectories, extracting per-node single-turn samples, and optimizing the orchestrator with GRPO-style training. On a 240-question test set spanning SGI-Reasoning and Scientists’ First Exam, SciOrch reaches 56.66% average accuracy, outperforming the strongest single commercial model by 3.74% and the strongest multi-agent baseline by 3.33%. It also attains the best accuracy on both SGI and SFE with less than half the API cost of typical multi-agent methods.

Static benchmarks often conflate memorization with reasoning, failing to capture the dynamic nature of world knowledge. We present LiveSearchBench, a scalable pipeline that automatically constructs snapshot-grounded benchmarks from knowledge graph differentials, in which every answer is unique and SPARQL-verifiable against a target snapshot. Unlike prior template-based approaches, our method leverages topology-aware synthesis to generate multi-constraint compositional questions with unique answers guaranteed via SPARQL validation. Evaluating 1,000 questions across three difficulty levels on both open-weight and frontier models, we expose a pronounced Recency Gap on facts that post-date a model’s training cutoff, particularly for compositional queries; retrieval dependence is thus an empirical property we measure, not a formal guarantee we claim. Through Oracle and Wikidata-as-corpus analyses, we disentangle dual failure modes: retrieval coverage limitations for single-hop questions and reasoning failures for compositional queries even with ground-truth evidence. LiveSearchBench provides both a rigorous construction methodology and a continuously updatable testbed for evaluating evidence integration under evolving knowledge.

Previous LLMs-based RL studies typically follow either supervised learning with high annotation costs, or unsupervised paradigms using voting or entropy-based rewards. However, their performance remains far from satisfactory due to the substantial annotation cost and issues such as model collapse or reward hacking. To address these issues, we introduce a new perspective inspired by cognitive learning theory and propose a novel approach called EasyRL. The core of EasyRL is to simulate the human cognitive acquisition curve by integrating reliable knowledge transfer from easy labeled data with a progressive divide-and-conquer strategy that tackles increasingly difficult unlabeled data. Specifically, we initialize a warm-up model using supervised RL with few-shot labeled data. This is followed by a divide-and-conquer pseudo-labeling strategy on difficult unlabeled data, combining consistency-based selection for low-uncertainty cases and reflection-based resolution for medium-uncertainty cases. Finally, difficulty-progressive self-training with iterative pseudo-labeling and RL further strengthens the model’s reasoning capability. EasyRL provides a unified self-evolving framework that facilitates data-efficient post-training of LLMs. Experimental results on mathematical and scientific benchmarks demonstrate that EasyRL, using only 10% of easy labeled data, consistently outperforms state-of-the-art baselines.

Large language model (LLM) routing assigns each query to the most suitable model from an ensemble. We introduce LLMRouterBench, a large-scale benchmark and unified framework for LLM routing. It comprises over 400K instances from 21 datasets and 33 models. Moreover, it provides comprehensive metrics for both performance-oriented and performance-cost trade-off routing, and integrates 10 representative routing baselines. Using LLMRouterBench, we systematically re-evaluate the field. While confirming strong model complementarity—the central premise of LLM routing—we find that many routing methods exhibit similar performance under unified evaluation, and several recent approaches, including commercial routers, fail to reliably outperform a simple baseline. Meanwhile, a substantial gap remains to the Oracle, driven primarily by persistent model-recall failures. We further show that backbone embedding models have limited impact, that larger ensembles exhibit diminishing returns compared to careful model curation, and that the benchmark also enables latency-aware analysis. All code and data are available at https://github.com/ynulihao/LLMRouterBench.

We propose Process-Aware Policy Optimization (PAPO), a method that integrates process-level evaluation into Group Relative Policy Optimization (GRPO) through **decoupled advantage normalization**, to address two limitations of existing reward designs. Outcome reward models (ORM) evaluate only final-answer correctness, treating all correct responses identically, and gradually lose the advantage signal as groups become uniformly correct. Rubric-based process reward models (PRM) offer richer supervision, differentiating quality among correct responses and restoring advantage diversity, but using their scores as rewards triggers verbosity-driven reward hacking and accuracy collapse. PAPO resolves both by composing the advantage from an outcome advantage A_out, derived from outcome reward r^out over all responses, and a process advantage A_proc, derived from process reward r^proc exclusively over the correct subset. A_out anchors correctness, and correct subset normalization on A_proc rules out verbosity-driven hacking while sustaining the gradient on all-correct groups where ORM exhausts. PAPO consistently outperforms ORM across five model configurations and six benchmarks (reaching 51.3% vs. 46.3% on OlympiadBench for Qwen2.5-7B while continuing to improve as ORM plateaus and declines), with analogous gains when composed with mainstream GRPO variants.

Deep research is an inherently challenging task that demands both breadth and depth of thinking. It involves navigating diverse knowledge spaces and reasoning over complex, multi-step dependencies, which presents substantial challenges for agentic systems. To address this, we propose FlowSearch, a multi-agent framework that actively constructs and evolves a dynamic structured knowledge flow to drive subtask execution and reasoning. FlowSearch is capable of strategically planning and expanding the knowledge flow to enable parallel exploration and hierarchical task decomposition, while also adjusting the knowledge flow in real time based on feedback from intermediate reasoning outcomes and insights. FlowSearch achieves competitive performance on both general and scientific benchmarks, including GAIA, HLE, GPQA and TRQA, demonstrating its effectiveness in multi-disciplinary research scenarios and its potential to advance scientific discovery. The code will be available.

Existing multi-LLM collaboration systems often encounter scalability challenges when integrating new LLMs and tasks, leading to suboptimal performance. To address this, we propose SMCS, a Scalable Multi-LLM Collaboration System designed to effectively coordinate multiple open-source LLMs. The system consists of two core components: a Retrieval-based Prior Selection (RPS) module, which dynamically selects the most suitable LLMs for each input, and an Exploration–Exploitation-Driven Posterior Enhancement (EPE) module, which fosters response diversity and selects high-quality outputs through a hybrid scoring mechanism. Experiments on eight mainstream benchmarks validate the effectiveness of our system: by integrating fifteen open-source LLMs, SMCS outperforms prevailing closed-source LLMs, e.g., GPT-4.1(+5.36%) and GPT-o3-mini(+5.28%) across multiple tasks. Remarkably, it even exceeds the average of best results on different datasets with open-source LLMs (+2.86%), significantly advancing the empirical performance frontier of open-source collaboration. The code is released at https://github.com/magent4aci/SMCS.

The rapid advancement of multimodal large language models (MLLMs) offers new opportunities for complex scientific challenges, yet their application in earth science—especially at the graduate level—remains underexplored due to a lack of benchmarks reflecting the depth and complexity of geoscientific reasoning. Existing datasets often rely on synthetic data or simple figure-caption pairs, failing to capture the nuanced reasoning required for real-world applications. To address this, we introduce MSEarth, a multimodal scientific dataset and benchmark curated from high-quality, open-access publications. Covering the five major spheres of Earth science—atmosphere, cryosphere, hydrosphere, lithosphere, and biosphere—MSEarth features over 289K figures with refined captions enriched by contextual discussions and reasoning from the original papers. The benchmark supports tasks such as scientific figure captioning, multiple choice questions, and open-ended reasoning, providing a scalable, high-fidelity resource for developing and evaluating MLLMs in scientific reasoning.

Multi-turn, long-horizon tasks are increasingly common for large language models (LLMs), but solving them typically requires many sequential model invocations, accumulating substantial inference costs. Here, we study cost-aware multi-turn LLM routing: selecting which model to invoke at each turn from a model pool, given a fixed cost budget. We propose MTRouter, which encodes the interaction history and candidate models into joint history–model embeddings, and learns an outcome estimator from logged trajectories to predict turn-level model utility. Experiments show that MTRouter improves the performance–cost trade-off: on ScienceWorld, it surpasses GPT-5 while reducing total cost by 58.7%; on Humanity’s Last Exam (HLE), it achieves competitive accuracy while reducing total cost by 43.4% relative to GPT-5, and these gains even carry over to held-out tasks. Further analyses reveal several mechanisms underlying its effectiveness: relative to prior multi-turn routers, MTRouter makes fewer model switches, is more tolerant to transient errors, and exhibits emergent specialization across models.Code: https://github.com/ZhangYiqun018/MTRouter

Evolution, the engine behind the survival and growth of life on Earth, operates through the population-based process of reproduction. Inspired by this principle, this paper formally defines a newly emerging problem: the population-based evolution of large language models (LLMs). We introduce a novel framework that starts with a population of parent LLMs and allows this population to evolve through four key operations: (i) crossover, merging the weights of different parents to create offspring LLMs, (ii) mutation, introducing small, random changes to model weights to foster diversity, (iii) selection, prioritizing high-performing models, and (iv) succession, transferring the learned experience from parent to offspring LLMs. With only 200 samples per new task, the LLM population evolves rapidly to adapt to the task at hand, without any gradients. Experiments on 12 datasets show that our framework consistently outperforms existing multi-LLM merging and adaptation methods, achieving relative performance gains of up to 54.8 over the best LLM in the initial population. Moreover, our framework allows for (i) the evolution of LLMs across multiple new tasks simultaneously, (ii) scaling effectively with populations of up to 40 LLMs, and (iii) even zero-shot generalization to unseen held-out tasks. Code: https://github.com/ZhangYiqun018/GENOME

Multi-step retrosynthetic planning is a fundamental challenge in organic chemistry, traditionally modeled as a combinatorial search problem guided by single-step prediction models. However, this search-centric paradigm often disconnects from the explicit chemical reasoning processes employed by human experts. In this paper, we propose R³ (Reinforced Reasoning Retrosynthesis), a novel framework that reformulates this task as end-to-end generative reasoning. Instead of traversing a search tree, R³ simulates the problem-solving logic of chemists to directly generate complete synthetic pathways. To achieve this, we initialize the model with domain knowledge and employ end-to-end Reinforcement Learning (RL) to optimize the entire planning policy. Experimental results on Retrobench show that R³ achieves a state-of-the-art Top-1 accuracy of 43.7%, demonstrating that generative reasoning offers a superior alternative to traditional search algorithms in solving complex retrosynthetic problems.

While scaling laws for large language models (LLMs) during pre-training have been extensively studied, their behavior under reinforcement learning (RL) post-training remains largely unexplored. This paper investigates the scaling behavior of Large Language Model (LLM) reinforcement learning post-training, focusing on mathematical reasoning. Through experiments across the Qwen2.5 series (0.5B to 72B), we characterize how model scale, data, and compute interact. Our analysis yields four key findings: 1. Larger models consistently demonstrate superior compute and data efficiency. 2. The relationship between model performance and training resources follows a predictive power-law across both base and instruction-tuned models. 3. RL learning efficiency exhibits a latent saturation trend with increasing model scale. 4. In data-constrained regimes, performance is primarily driven by the total volume of training data rather than sample uniqueness. These results offer practical guidelines for scaling reasoning capabilities through reinforcement learning post-training.

Reinforcement learning (RL) has emerged as a powerful post-training paradigm for enhancing the reasoning capabilities of large language models (LLMs). However, reinforcement learning for LLMs faces substantial data scarcity challenges, including the limited availability of high-quality external supervision and the constrained volume of model-generated experience. These limitations make data-efficient reinforcement learning a critical research direction. In this survey, we present the first systematic review of reinforcement learning for LLMs under data scarcity. We propose a bottom-up hierarchical framework built around three complementary perspectives: the data-centric perspective, the training-centric perspective, and the framework-centric perspective. We develop a taxonomy of existing methods, summarize representative approaches in each category, and analyze their strengths and limitations. Our taxonomy aims to provide a clear conceptual foundation for understanding the design space of data-efficient RL for LLMs and to guide researchers working in this emerging area. We hope this survey offers a comprehensive roadmap for future research and inspires new directions toward more efficient and scalable reinforcement learning post-training for LLMs.

We introduce MinerU2.5, a 1.2B-parameter document parsing vision-language model that achieves state-of-the-art recognition accuracy while maintaining exceptional computational efficiency. Our approach employs a coarse-to-fine, two-stage parsing strategy that decouples global layout analysis from local content recognition. In the first stage, the model performs efficient layout analysis on downsampled images to identify structural elements, circumventing the computational overhead of processing high-resolution inputs. In the second stage, guided by the global layout, it performs targeted content recognition on native-resolution crops extracted from the original image, preserving fine-grained details in dense text, complex formulas, and tables. To support this strategy, we developed a comprehensive data engine that generates diverse, large-scale training corpora for both pretraining and fine-tuning. Ultimately, MinerU2.5 demonstrates strong document parsing ability, achieving state-of-the-art performance on multiple benchmarks, surpassing both general-purpose and domain-specific models across various recognition tasks, while maintaining significantly lower computational overhead.

Multi-agent systems have emerged as a promising approach for enhancing the reasoning capabilities of large language models in complex problem-solving. However, current MAS frameworks are limited by poor flexibility and scalability, with underdeveloped optimization strategies. To address these challenges, we propose ReSo, which integrates task graph generation with a reward-driven two-stage agent selection process. The core of ReSo is the proposed Collaborative Reward Model, which can provide fine-grained reward signals for MAS cooperation for optimization. We also introduce an automated data synthesis framework for generating MAS benchmarks, without human annotations. Experimentally, ReSo matches or outperforms existing methods. ReSo achieves 33.7% and 32.3% accuracy on Math-MAS and SciBench-MAS SciBench, while other methods completely fail. The code and data are available at Reso.

Survey paper plays a crucial role in scientific research, especially given the rapid growth of research publications. Recently, researchers have begun using LLMs to automate survey generation for better efficiency. However, the quality gap between LLM-generated surveys and those written by human remains significant, particularly in terms of outline quality and citation accuracy. To close these gaps, we introduce SURVEYFORGE, which first generates the outline by analyzing the logical structure of human-written outlines and referring to the retrieved domain-related articles. Subsequently, leveraging high-quality papers retrieved from memory by our scholar navigation agent, SURVEYFORGE can automatically generate and refine the content of the generated article. Moreover, to achieve a comprehensive evaluation, we construct SurveyBench, which includes 100 human-written survey papers for win-rate comparison and assesses AI-generated survey papers across three dimensions: reference, outline, and content quality. Experiments demonstrate that SURVEYFORGEcan outperform previous works such as AutoSurvey.

The scientific research paradigm is undergoing a profound transformation owing to the development of Artificial Intelligence (AI). Recent works demonstrate that various AI-assisted research methods can largely improve research efficiency by improving data analysis, accelerating computation, and fostering novel idea generation. To further move towards the ultimate goal (i.e., automatic scientific research), in this paper, we introduce Dolphin, a closed-loop LLM-driven framework to enhance the automation level of scientific research. Dolphin first generates novel ideas based on feedback from previous experiments and relevant papers ranked by the topic and task attributes. Then, the generated ideas can be implemented using a code template refined and debugged with the designed exception-traceback-guided local code structure. Finally, Dolphin automatically analyzes the results of each idea and feeds the results back to the next round of idea generation. Experiments are conducted on the benchmark datasets of different topics and a subset of MLE-bench. Results show that Dolphin can continuously improve the performance of the input topic in a loop. We highlight that Dolphin can automatically propose methods that are comparable to the state-of-the-art in some tasks such as 3D point classification.