This study compares programmatic tool calling (PTC) with native JSON calls across 14 models on BFCL v4, showing a 10.6% accuracy boost for PTC, demonstrating robustness and scalability.

Ishan Patel, Sahil Sen, Elias Lumer et al.

Proposes a reference-free, multi-metric framework using LLM judges to evaluate conversational benchmarks' consistency, complexity, and coverage, validated via synthetic and manual benchmarks.

Noam Koren, Roy Bar-Haim, Abigail Goldsteen

Introduces GB/T-Bench and GB/T-Reviewer, leveraging hierarchical error taxonomy and multi-agent collaboration to significantly improve rule-based national standard document review.

Tao Wang, Qihao Yang, Rongjiao Liang et al.

HiLP method introduces higher-level abstract latents to enhance long-horizon reasoning in language models.

Chang Shi, Tim Pearce, Manan Tomar et al.

Proposes Spoken Function Calling (SFC), leveraging structured rules to enhance large audio language models' semantic understanding beyond traditional closed-set SLU.

Yuezhang Peng, Yuxin Liu, Changfeng Gao et al.

EuroExec benchmark reveals frontier LLMs achieve only 56.9% solve rate on European executive tasks, far below expert performance.

Pau Arnal, Khaled Denfir, Danylo Smahliuk et al.

Using Qwen2.5-0.5B, recursive training raised EO GAP 13.18→19.38 while PPL improved 16.07→10.55.

Irina Proskurina, Antoine Gourru, Julien Velcin

VetScore integrates citation support and harm potential scoring for veterinary long-form QA verification, improving trustworthiness.

Ivan Kartáč, Jan Tovarys, Mateusz Lango et al.

This study analyzes SFT and RL in multi-task learning, revealing RL induces sparse, orthogonal parameter updates, reducing task interference.

Kejian Zhu, Zhuoran Jin, Shangqing Tu et al.

Proposes ICO, an iterative context optimization framework, achieving 74.6% success in semantic-shift jailbreaks with high-quality contexts.

Hujian Zhu, Yihao Huang, Felix Juefei-Xu et al.

Distinguishing emulation and estimation tasks, base models excel in response generation, fine-tuned models in direct distribution prediction.

Seth Grief-Albert, Jessica Bo, Difan Jiao et al.

Introduces PCTree, leveraging parent-conditioned scoring to boost semi-autoregressive decoding speed by up to 29.5%.

Zixian Li, Tong Li, Chi Xie et al.

Proposes Cloud-ScPO, leveraging hidden-state geometry for semi-supervised preference optimization, achieving up to 4.49% accuracy gain on GSM8K.

Yuzhou Liu, Xiyang Hu

Pruned BPE introduces post-training visibility pruning and token reallocation, reducing encoded length by ~0.3% while preserving vocabulary size.

Kenny Shao

OoO-Spec employs asynchronous semantic speculation, achieving up to 5.34× speedup in tool calls with a single trained sidecar model.

Zhiheng Zhang, Mujie Xu, Feiyu Sun et al.

AdaMTP uses entropy-based boundary detection and dynamic masking to improve multi-token prediction accuracy and speed.

Ziqiang Cui, Han Shi, Bowei He et al.

Introduced COMPINT suite to address session constraint loss in context compaction, achieving 90% retention.

Zhiqi Wang, Yichi Zhang, Dongwon Lee et al.

Data Turnstile framework generates high-quality function-calling data, boosting Qwen3-0.6B accuracy to 75.9%.

Goutham Ramakrishnan, Megha Sharma

DuPLeR employs dual-path structural reasoning combined with multimodal LLM priors to improve few-shot knowledge graph completion, achieving significant performance gains.

Jinlan Liu, Zhiying Tu, Yongchao Xing et al.

Metis integrates native memory into foundation models, using a parametric memory state and self-supervised training to enhance long-term reasoning.

Zeyu Zhang, Ziliang Guo, Yihang Sun et al.