AI research papers.

AI agents are demonstrating increasingly strong offensive and defensive cybersecurity capabilities in the real world.

Long-horizon agents accumulate progress in persistent task state, yet repeatedly process interaction histories describing how that progress was achieved.

Evaluation for automatic crisis reporting should measure what practitioners need, not what is convenient to compare.

Post-training is crucial for improving text-to-image models, yet existing pipelines typically rely on manually curated training data and fixed training distributions.

We introduce LensErase, a general-purpose mask-based Erase Anything foundation model designed to cover a broad spectrum of real-world erase scenarios.

Effective collaboration among large language model (LLM) agents in multi-agent systems requires sharing relevant evidence, requesting missing information, and incorporating teammates' contributions into decisions.

Textual reasoning before acting has improved vision-language-action (VLA) models, but at the cost of latency: every action conditioned on the reasoning text must wait for autoregressive decoding.

Recent video super-resolution (VSR) models achieve efficient one-step generation through diffusion distillation.

Complex and long-horizon agent tasks often rely on frontier models deployed in the cloud to achieve strong performance.

For video generation in dynamic world modeling, observed static and dynamic content should be efficiently reused to provide conditions for repeated generation at different times and viewpoints.

World Action Models (WAMs) use generative visual representations to condition robot actions, yet how these representations organize action-relevant information remains poorly understood.

Knowledge distillation (KD) frameworks place teacher and student on the same GPUs, so per-GPU memory limits the teachers that fit.

Reliable video reasoning requires models to produce correct answers and expose supporting visual evidence.

Reinforcement learning with verifiable rewards (RLVR) for multi-turn agents relies on sparse terminal rewards that say little about individual steps.

Cowork agents navigate dependencies across tools and files while adapting to changing user requirements and edits to shared workspaces.

AI systems now write long and difficult mathematical proofs, and checking their step-by-step validity is becoming a bottleneck.

Long-horizon, semi-verifiable agentic tasks often rely on rubric-based evaluation, but the evidence needed to assess these tasks can be spread across the environment state and the execution trace.

A recommendation request on Reddit draws several replies, each naming different titles.

Edge-cloud collaboration combines the lower inference overhead and greater data locality of edge-side execution with the stronger capabilities of cloud-hosted models.

Diffusion large language models (dLLMs) are appealing for code generation because they can update many tokens in parallel, yet many are optimized for sequence synthesis rather than edit prediction.

Agent Skills improve LLM agents by making task knowledge reusable across sessions.

Long-context training of large language and multimodal models couples structured attention masks (document, prefix-LM, share-question, sliding-window, block-sparse) with multi-node context parallelism, making the mask a…

Recent advances in vision-language models (VLMs) have substantially improved the capabilities of GUI agents, while memory augmentation has emerged as a promising approach for reusing knowledge from past interactions.

Large language models have evolved from single-turn chatbots into interactive agents capable of long-horizon reasoning.

The design of attention mechanisms in long-context large language models faces an ``impossible triangle'' among KV cache size, training and prefill FLOPs, and model quality.

We introduce Fish Audio Drama 3, an open-source speech generation model that turns text into directed performances, built on a 35B-parameter mixture-of-experts backbone that activates 3B parameters per token.

Test-time scaling has enabled LLMs to solve increasingly complex reasoning, discovery, and optimization tasks.

Scaling interactive environments is important for training long-horizon agents through reinforcement learning.

Long-horizon agents often struggle to recover from failures whose causes lie several steps earlier, beyond what local reflection or backtracking can resolve.

Deep research (DR) agents address open-ended information needs through multi-step reasoning, long-horizon web search, and evidence synthesis.

Training-free diffusion-based compression methods repurpose the generative priors of a diffusion or a flow model to compress data without retraining them as task-specific codecs.

We present SWE-MISBEHAVIOR, the first taxonomy and evaluation framework for diagnosing boundary-violating misbehaviors in coding agents.

Benchmarks are essential for evaluating and improving large language models (LLMs).

Autonomous research agents must acquire and reason over evidence within a limited budget.

Generated images often have coherent layouts and visual content, yet local errors in small text and complex characters still limit their readability and practical use.

A robot following a multi-step instruction needs to act on the right objects in the right order, even as the scene changes.

Game-theoretic benchmarks claim that large language model (LLM) strategic decision-making varies across game types, interaction structures, and capability dimensions.

City-scale 3D grounding is commonly formulated as candidate ranking, which requires precomputed instance proposals and becomes inefficient in large outdoor scenes.

Few-step distillation reduces the sampling cost of diffusion and flow models, but existing approaches often require auxiliary score models or specialized architectures.

Terminal agents have shown strong abilities in complex coding tasks.

Tool-call judges evaluate argument values proposed by LLM agents.

Vision-language-action (VLA) policies excel at generalist robotic manipulation but generate actions without explicitly modeling their environmental consequences.

System One models choose an agent's next action by scoring a supplied set of options in a single pass.

Reinforcement learning (RL) post-training improves robot foundation models (RFMs) beyond their demonstrations, but real-world deployment demands more: learning new tasks efficiently while retaining previously acquired be…

Multimodal LLMs are typically trained by first aligning a vision encoder with a language model via projector training on large image-caption datasets, followed by joint training of the full multimodal model.

3D part assembly aims to reconstruct a complete shape from a set of unassembled parts by predicting a rigid SE(3) transformation for each part.

Controlling an acting LLM agent means deciding, at each step of a long-horizon task, whether to trust its next action, abstain, or hand off to a human.

Visual agents must turn video observations into timely decisions to track progress and choose what to do next.

Combining pretrained models is a common strategy for building AI systems, whether by aligning a vision encoder with a language model or merging fine-tuned checkpoints.

Online decision-making faces a tension between efficient inference and long-horizon reasoning.