ConceptARC at Scale introduces a generator that produces thousands of ARC-style tasks focused on specific spatial and semantic concepts, expanding the original benchmark by tenfold. The new dataset, ConceptARC-100, contains 1,600 manually reviewed tasks across 16 concepts, each with 100 tasks. Evaluation of large reasoning models on this dataset reveals that they reliably solve certain concepts, struggle with others, and develop internal representations that cluster by transformation operations such as recoloring, removal, or translation.
By Philipp Mondorf, Adrien Deli\`ege, Solim LeGris, Claas Beger, Ryan Yi, Barbara Plank, Brenden M. Lake, Todd M. Gureckis
PotARCin expands the ARC benchmark by evaluating abstract reasoning across five dimensions—Definition, Classification, Constrained Generation, Editing, and Inversion—using programmatic generation of new task instances. The study shows a 25‑52 percentage‑point performance gap between standard ARC evaluation and PotARCin, and reveals that multi‑dimensional assessment can reorder models that appear equivalent under single‑metric accuracy. Additionally, a new held‑out set, P‑ARC, demonstrates low model accuracy (1‑8%) across all dimensions, highlighting the need for more comprehensive tests of abstract reasoning.
By Claas Beger, Ryan Yi, Melanie Mitchell
arXiv:2605. 28742v2 Announce Type: replace Abstract: Language models can use verifiable rewards to improve at a wide variety of reasoning tasks.
By Linas Nasvytis, Simon Jerome Han, Ben Prystawski, Satchel Grant, Noah D. Goodman, Judith E. Fan
arXiv:2606. 07410v1 Announce Type: cross Abstract: The emergence of "Aha moments" in large language models, particularly DeepSeek-R1-0120, has raised the question of whether these systems genuinely reason or merely imitate the appearance of reasoning.
By Yuxiang Chen, Jun Wang
arXiv:2511. 18121v2 Announce Type: replace-cross Abstract: While Multimodal Large Language Models (MLLMs) excel on benchmarks, their processing paradigm differs from the human ability to integrate visual information.
By Ming Zhong, Yuanlei Wang, Liuzhou Zhang, Ruichuan An, Renrui Zhang, Hao Liang, Ming Lu, Ying Shen, Wentao Zhang
LiveMathematicianBench is a dynamic multiple‑choice benchmark for research‑level mathematical reasoning, built from recent arXiv papers published after model training cutoffs. It introduces a thirteen‑category logical taxonomy of theorem types and uses a proof‑sketch‑guided distractor pipeline to create plausible but invalid answer choices, enhancing sensitivity to genuine reasoning. Evaluation shows current large language models perform poorly, with the best model scoring 43.5% overall and only 17.6% under substitution‑resistant conditions, indicating the benchmark’s difficulty and realism.
By Linyang He, Qiyao Yu, Hanze Dong, Baohao Liao, Xinxing Xu, Micah Goldblum, Jiang Bian, Nima Mesgarani
arXiv:2605. 03344v2 Announce Type: replace-cross Abstract: Retrieval-augmented generation (RAG) has proven effective for knowledge-intensive tasks, but is widely believed to offer limited benefit for reasoning-intensive problems such as math and code generation.
By Negar Arabzadeh, Wenjie Ma, Sewon Min, Matei Zaharia
arXiv:2606. 26530v1 Announce Type: cross Abstract: The Abstraction and Reasoning Corpus (ARC;~\citealp{chollet2019measure}) contains tasks that require summarizing patterns from limited grid samples and predicting output grids.
By Yuxuan Yang, Feiyang Li, Yile Wang
InternBootcamp is an open‑source framework that offers over 1,000 domain‑diverse task environments for large language model (LLM) reasoning research. It introduces Bootcamp‑Eval, an automatically generated benchmark for comprehensive performance assessment. Experiments show that training on InternBootcamp significantly improves reasoning performance, with a 32B model achieving state‑of‑the‑art results on Bootcamp‑Eval and other established benchmarks, demonstrating that scaling the number of training tasks yields consistent gains.
By Peiji Li, Jiasheng Ye, Yongkang Chen, Linyang Li, Yichuan Ma, Zijie Yu, Ganqu Cui, Haozhan Li, Jiacheng Chen, Chengqi Lyu, Wenwei Zhang, Qipeng Guo, Dahua Lin, Bowen Zhou, Kai Chen
The paper introduces a two‑step test‑time training protocol, Embed‑TTT, for Vision ARC (VARC) that first fine‑tunes only the task embedding and then fine‑tunes the backbone. This approach consistently produces task embeddings that better align with the underlying rules, improves retrieval and linear probing, and recovers the geometric structure of parametric rules. Even fine‑tuning only the tiny embedding component solves a significant portion of ARC‑AGI‑1, ConceptARC, and Mini‑ARC tasks, while the full two‑step pipeline further enhances performance and demonstrates compositional rule interpolation.
By Adrien Deli\`ege, Claas Beger, Marc Van Droogenbroeck, Melanie Mitchell
arXiv:2508. 09883v2 Announce Type: replace-cross Abstract: Large language models (LLMs) demonstrate remarkable reasoning capabilities in tasks such as algorithmic coding and mathematical problem-solving.
By Xiaojun Wu, Xiaoguang Jiang, Huiyang Li, Jucai Zhai, Dengfeng Liu, Qiaobo Hao, Huang Liu, Zhiguo Yang, Ji Xie, Ninglun Gu, Jin Yang, Kailai Zhang, Yelun Bao, Jun Wang
arXiv:2511. 17731v2 Announce Type: replace-cross Abstract: Chain-of-Thought (CoT) prompting has proven remarkably effective for eliciting complex reasoning in large language models (LLMs).
By Lingxiao Li, Yifan Wang, Xinyan Gao, Chen Tang, Xiangyu Yue, Chenyu You