HuggingFace

Hugging Face introduces transformers-mic, a Model Integration Contract (MIC) standard that packages tiny published fixtures, runnable usage paths, and human-reviewed baselines to catch model regressions on cheap CPU runners before they reach the Transformers main branch.

Reviso is an agentic video captioning framework that uses hierarchical Context-Focus-Points planning and active perception to prioritize key visual evidence, trained with GRPO and specialized rewards, outperforming open-source baselines on four video captioning benchmarks.

The paper introduces Library of Layers (LoL), a transformer architecture that dynamically reorders its layers at inference time, unifying parameter reuse and sparsity tradeoffs across dense, depth-recurrent, and MoE baselines while enabling interpretable routing across reasoning, multilingual, and safety tasks.

FLAMIA 提出了一族面向 NVIDIA Blackwell GPU 的混合精度注意力内核，利用 MXFP8 与 NVFP4/FP8 路径，在避免因果信息泄漏的同时加速 Transformer 训练，在 B200/B300 上实现约 1.06–1.42 倍的前向与反向注意力加速。

CoworkOS is a long-horizon agent harness that separates strategy from execution, enabling human-in-the-loop collaboration through shared artifacts and sustained automated post-training that improved Qwen3.6-35B-A3B's average benchmark score from 33.19 to 42.95 over five RSI-driven rounds.

CUA-Master 提出嵌套双环递归改进框架，通过开放世界语料、自催化任务合成与反馈引导的快照回滚重试机制持续提升计算机使用代理（CUA）性能，其 27B 模型在 OSWorld Verified 等基准上取得竞争力表现。

Lab2Sim introduces a benchmark of 218 science experiment videos with evidence-grounded specifications, along with EduVideo2Code, which encodes procedures as state machines to guide verified code generation for interactive virtual experiments, improving apparatus fidelity by 24.15 points.

The paper introduces the Collaborative Intelligence Benchmark, which evaluates AI agents as teammates across 240 tasks in 8 teamwork dimensions, finding that even top models score at most 82.2/100 and that general capability does not guarantee strong teamwork.

DR Pixel is an open-source end-to-end multimodal deep research agent that retrieves and reasons directly over rendered document page images instead of OCR-extracted text, introducing counterfactual credit reweighting for agentic supervised finetuning. The approach outperforms OCR-based baselines on DRPixelBench across models like GPT-5.4 and Gemini-3.5-Flash, boosting retrieval recall by ~15 points and accuracy by 10+ points after training.

An empirical study finds that continual harness self-evolution of coding agents across multiple software engineering tasks causes severe catastrophic forgetting, with SWE-Bench performance dropping from 94.0% to 32.0% after evolving on FeatureBench, and offers insights for future continual agent-harness evolution.

PhotoGuider 提出一种 measure-then-frame 方法，利用共享的构图感知 token 进行 360° 全景图的观看方向选择与取景，并通过微调 MLLM 生成以测量结果为条件的构图解释，在两项 2D 裁剪基准上取得最高 mIoU，并在 120 人盲测中以 53.2% 的偏好率排名第一。

TaoMate-FVSR proposes an audio-guided facial video restoration framework that combines reference-guided video flow modeling with audio synchronization to recover lip, dental, and facial details in AI-generated videos, achieving state-of-the-art perceptual similarity and synchronization on talking-head benchmarks.

LeanGuard is a runtime that verifies every tool call from language-model agents against Lean4 policies with temporal logic semantics, cutting forbidden calls and prompt-injection success while enabling proof-guided repair of denied actions.

The paper proposes Avatar Intelligence, a paradigm of persistent personal agents that represent users' goals, values, and constraints through sustained decision delegation, and introduces the Agentic Oversight Index (AOI) to evaluate delivered value against operational cost and oversight burden, outlining a research agenda covering identity-continuous personalization, Rehearsal Engineering, and safety mechanisms.

This paper evaluates frozen video representations and finds that latent prediction errors in video world models reflect appearance and motion rather than reliably measuring physical surprise, presenting a framework that separates representation content from error signals across IntPhys2, Physion++, and Kubric benchmarks.

CrossSWE introduces a multi-repository benchmark of 171 real-world instances showing that current LLM coding agents struggle with cross-repository consistency, with the strongest model (GPT-6.1 Sol) reaching only 58.48% Pass@1 and cross-repository inconsistencies accounting for 44.2% of failures.

SWE-Trail is a history-guided trajectory collector for coding agents that uses earlier rollouts to steer where collection effort is spent without exposing that history to the agent, avoiding train-test mismatch. Trained on its trajectories, a 124B Ling-3.0-flash-base student gains 2.33 and 1.60 percentage points in resolution on SWE-bench Multilingual and SWE-bench Pro over independent sampling.

The paper introduces Tutorial Agentic Learning (TAL), a nonparametric continual-learning paradigm that lets LLM agents learn from instructional videos by building a structured tutorialbook memory and reasoning-linked table-of-contents retrieval, plus a VidLearn chess-video QA benchmark where TAL significantly outperforms vanilla MLLMs.

A report proposing a unified taxonomy and an agentic literature system for LLM-driven computer graphics, analyzing trends and open challenges in human-agent collaboration and long-horizon planning.

NorthCUA technical report presents an integrated system that scales computer-use agents via an Experience Flywheel, recursive self-improvement, staged learning, and reinforcement learning, achieving state-of-the-art open-source results on GUI grounding, desktop execution, and web interaction with its Nano, Mini, and Pro model family.