GSM-Plus-BN evaluates LLMs on Bengali math reasoning with perturbations; GPT-OSS-20B achieves 96.08% accuracy on seed questions under Standard Prompting.

Bidyarthi Paul, Nahida Jannat Mayouree, Md. Asif Karim et al.

ESFP benchmark measures epistemic stance flexibility in large language models under different prompts.

Binwen Liu, Yilin Ren

EcoSpec optimizes MoE model inference by considering expert activation costs, achieving a 1.62x decoding speedup.

Jincheng Xie, Runheng Liu, Heyan Huang et al.

RefineEvo employs planning-guided heuristic evolution and bidirectional experience pools to enhance combinatorial optimization.

Yang Wu, Junran Pan, Yifan Zhang et al.

ProgramTab leverages Python-guided code generation to enhance large table reasoning, outperforming baselines on WikiTQ and TabFact.

Pei Guo, Enjie Liu, Yunzhi Tan et al.

Proposed QFES task and QFESum dataset; combined RAT and SHC frameworks outperform baselines in event-focused summarization.

Chenyu Hu, Bang Wang

Introduced QIMG-7 benchmark and SATR method to enhance multimodal RAG reliability in polluted settings, improving balanced score by 11.7 points.

Saadeldine Eletter, Owais Aijaz, Preslav Nakov

A ReAct retrieval and fine-tuned ensemble reaches 0.151 class-balanced Brier on 404 M&A deals, outperforming markets, XGBoost, and frontier LLMs.

Hinal Jajal, Michal Mucha, Charles Sweat et al.

Test-Time Scaling (TTS) significantly improves reasoning in small VLMs on multilingual visual MCQ, achieving 84.1% accuracy in the best configuration.

Spiros Baxevanakis, Peng-Jian Yang

Procrustes-conditioned joint end-to-end Top-K sparse autoencoder extracts universal features across independent BERT seeds, achieving Pearson r≥0.70.

Bendegúz Váradi, Zoltán Kmetty

DeepSearch-Evolve achieves self-distillation in DeepSearch-World, reaching 93.4% on HotpotQA.

Xinyu Geng, Xuanhua He, Sixiang Chen et al.

DeLS-Spec improves inference speed and acceptance length by combining long and short contexts.

Hong-Kai Zheng, Piji Li

Pluralis v0.1 introduces a culture-first multimodal evaluation framework with 6,448 prompts, exposing VLM blind spots in cultural alignment.

Alicia Parrish, Rajat Shinde, Sanket Badhe et al.

This study systematically compares BPE and Unigram-LM tokenizers on fixed 165-token chemical bases, revealing near-disjoint vocabularies and emphasizing tokenizer choice as a key model design decision.

Hunter Heidenreich

Audex, built on Nemotron-Cascade-2-30B-A3B, unifies audio-text modeling with a single Transformer decoder, achieving state-of-the-art performance in audio understanding, speech recognition, translation, and generation, while maintaining text reasoning.

Zhifeng Kong, Sang-gil Lee, Jaehyeon Kim et al.

Introduces MTEB-BR, a benchmark with 22 native Portuguese tasks, evaluating 93 models using rigorous statistical analysis to distinguish capability tiers.

Tardelli Ronan Coelho Stekel

dOPSD enhances reasoning in diffusion language models by self-distillation, excelling on Dream and LLaDA datasets.

Phuong Tuan Dat, Qi Li, Xinchao Wang

CrossHallu evaluates internal signals' cross-lingual and cross-domain generalization in LLMs, showing most models transfer hallucination signals effectively.

Aisha Alansari, Malak Alkhorasani, Hamzah Luqman

AdaptiveSpec boosts inference speed by 56% and recovers 93% accuracy by dynamically adjusting draft tree shapes and lossy verification.

Oszkár Urbán, Young D. Kwon, Stylianos I. Venieris et al.

DCCD combines document and token confidence to improve conflict resolution in multi-document QA, outperforming baselines on DRQA.

Raymond Li, Md Tawkat Islam Khondaker, Amirhossein Abaskohi et al.