Yuanfu Wang, Zhixuan Liu, Xiangtian Li, Chaochao Lu, Chao Yang · Feb 12, 2026 · Citations: 0
DemonstrationsMathCoding
The prevailing paradigm for training large reasoning models--combining Supervised Fine-Tuning (SFT) with Reinforcement Learning with Verifiable Rewards (RLVR)--is fundamentally constrained by its reliance on high-quality, human-annotated…
This dependency incurs significant data-collection costs, risks embedding human cognitive biases, and confines the reinforcement learning stage to objectively assessable domains like mathematics and coding, leaving a wide range of…
Ailin Huang, Ang Li, Aobo Kong, Bin Wang, Binxing Jiao, Bo Dong · Feb 11, 2026 · Citations: 0
Pairwise PreferenceMathCoding
We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency.
Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and…
Avinash Reddy, Thayne T. Walker, James S. Ide, Amrit Singh Bedi · Feb 8, 2026 · Citations: 0
Automatic MetricsMath
Across structured reasoning benchmarks, DCCD improves strict structured accuracy by up to +24 percentage points over standard constrained decoding (e.g., 15.2\% to 39.0\% on GSM8K with a 1B model), and enables smaller model pairs to match…
Tao Liu, Jiafan Lu, Bohan Yu, Pengcheng Wu, Liu Haixin, Guoyu Xu · Feb 5, 2026 · Citations: 0
Automatic MetricsMathCoding
Although current methods perform well on benchmarks like BIRD and Spider, they struggle with complex reasoning, domain knowledge, and hypothetical queries, and remain costly in enterprise deployment.
Experimental results demonstrate that IESR achieves state-of-the-art performance on the complex reasoning benchmark LogicCat (24.28 EX) and the Archer dataset (37.28 EX) using only compact lightweight models without fine-tuning.
Yuanjie Lyu, Chengyu Wang, Lei Shen, Jun Huang, Tong Xu · Jan 30, 2026 · Citations: 0
Rubric RatingSimulation EnvMath
Small LLMs often struggle to match the agentic capabilities of large, costly models.
While reinforcement learning can help, progress has been limited by two structural bottlenecks: existing open-source agentic training data are narrow in task variety and easily solved; real-world APIs lack diversity and are unstable for…
Further reliability evaluations on factual knowledge, question answering, and missing-premise scenarios confirm the strong generalization and robustness of PIR.
Nuoya Xiong, Yuhang Zhou, Hanqing Zeng, Zhaorun Chen, Furong Huang, Shuchao Bi · Jan 8, 2026 · Citations: 0
Expert VerificationMathCoding
Empirically, across both Llama-3 and Gemma-2 families and diverse benchmarks spanning mathematical reasoning, code generation, and instruction following, FusionRoute outperforms both sequence- and token-level collaboration, model merging,…
We evaluated seven expansion ratio configurations using comprehensive benchmarks assessing factual knowledge, mathematical reasoning, language comprehension, instruction-following, and truthfulness.
Qihao Liu, Luoxin Ye, Wufei Ma, Yu-Cheng Chou, Alan Yuille · Dec 18, 2025 · Citations: 0
Pairwise PreferenceAutomatic MetricsMath
Across various mathematical benchmarks, the method delivers consistent gains over strong baselines with standard RL post-training.
The modular discriminator also enables flexible reward shaping for objectives such as teacher distillation, preference alignment, and mathematical proof-based reasoning.
Ngoc Bui, Shubham Sharma, Simran Lamba, Saumitra Mishra, Rex Ying · Dec 3, 2025 · Citations: 0
Automatic MetricsMath
Across mathematical reasoning (GSM8K, MATH-500, AIME24), procedural generation (LongProc), conversational long-memory benchmarks (LongMemEval), and long-context understanding (LongBenchV2 and SCBench), TRIM-KV consistently outperforms…
Qualitative analyses further reveal that learned retention scores align with human intuition, naturally recovering heuristics such as sink tokens, sliding windows, and gist compression without explicit design.
Salman Rahman, Sruthi Gorantla, Arpit Gupta, Swastik Roy, Nanyun Peng, Yang Liu · Dec 2, 2025 · Citations: 0
Critique EditAutomatic MetricsMath
On ProcessBench, a benchmark for identifying erroneous steps in mathematical reasoning, PRMs trained on step-level self-consistency data achieve 67.5 F1, surpassing reference-guided training with ground-truth access (66.4 F1) and GPT-4o as…
When deployed as reward signals in RL training with Qwen2.5-Math-7B, our best PRM achieves 47.4% average accuracy across six mathematical reasoning benchmarks, outperforming ground-truth-based RLVR (43.9%).
Long Lian, Sida Wang, Felix Juefei-Xu, Tsu-Jui Fu, Xiuyu Li, Adam Yala · Nov 24, 2025 · Citations: 0
Automatic MetricsMath
Across six challenging math reasoning benchmarks, ThreadWeaver trained on top of Qwen3-8B achieves performance on par with cutting-edge sequential reasoning models (79.9% on AIME24 and 71.9% on average) while delivering up to 1.53x speedup…
Yijia Fan, Jusheng Zhang, Kaitong Cai, Jing Yang, Chengpei Tang, Jian Wang · Nov 17, 2025 · Citations: 0
Automatic MetricsMathCoding
To address this, we introduce the Dynamic Auction-based Language Agent (DALA), a novel framework that treats communication bandwidth as a scarce and tradable resource.
Extensive and comprehensive experiments demonstrate that our economically-driven DALA achieves new state-of-the-art performance across seven challenging reasoning benchmarks, including 84.32% on MMLU and a 91.21% pass@1 rate on HumanEval.
Rui Jia, Min Zhang, Fengrui Liu, Bo Jiang, Kun Kuang, Zhongxiang Dai · Nov 8, 2025 · Citations: 0
Automatic MetricsMath
To address these challenges, we propose EduAgentQG, a multi-agent collaborative framework for personalized mathematics question generation with explicit diversity and objective-aware evaluation.
Experiments show that EduAgentQG consistently outperforms COT, COT_N, ReAct, and EQPR in diversity, Objective Consistency, and Win Rate.
David Acuna, Chao-Han Huck Yang, Yuntian Deng, Jaehun Jung, Ximing Lu, Prithviraj Ammanabrolu · Nov 7, 2025 · Citations: 0
Pairwise PreferenceMath
We introduce a framework able to synthesize vision-centric problems spanning diverse levels of complexity, and the resulting dataset with over 1M high-quality problems including: reasoning traces, preference data, and instruction prompts…
Remarkably, finetuning Qwen2.5-VL-7B on our data outperforms existing open-data baselines across evaluated vision-centric benchmarks, and our best configurations match or surpass strong closed-data models such as MiMo-VL-7B-RL on Vstar…
Ali Asgarov, Umid Suleymanov, Aadyant Khatri · Oct 31, 2025 · Citations: 0
Automatic MetricsMathCoding
We introduce SIGMA (Search-Augmented On-Demand Knowledge Integration for AGentic Mathematical reAsoning), a unified framework that orchestrates specialized agents to independently reason, perform targeted searches, and synthesize findings…
When evaluated on challenging benchmarks such as MATH500, AIME, and PhD-level science QA GPQA, SIGMA consistently outperforms both open- and closed-source systems, achieving an absolute performance improvement of 7.4%.
This paper proposes Group-relative Implicit Fine-Tuning (GIFT), a reinforcement learning framework for aligning large language models (LLMs) that unifies on-policy optimization with implicit preference learning.
Results show that GIFT converges faster, generalizes better with reduced overfitting, and outperforms GRPO on mathematical reasoning benchmarks (GSM8K, MATH, AIME) as well as generation tasks' evaluations (AlpacaEval and Arena-Hard).
Shangqing Tu, Yaxuan Li, Yushi Bai, Lei Hou, Juanzi Li · Oct 9, 2025 · Citations: 0
Llm As JudgeAutomatic MetricsMathCoding
Our method features a specialized judge model trained with out-of-distribution data (AIME 2022, AIME 2023, and MATH 500) using oversampling techniques to accurately predict answer equivalence from partial reasoning traces, achieving 0.7072…
Comprehensive evaluations across three challenging benchmarks (AIME 2024, AIME 2025, and GPQA) and multiple reasoning models demonstrate that DeepPrune achieves remarkable token reduction of 65.73%--88.50% compared to conventional consensus…
