model releases

Two-dimensional early exit optimisation of LLM inference

arXiv:2604.18592v1 Announce Type: cross Abstract: We introduce a two-dimensional (2D) early exit strategy that coordinates layer-wise and sentence-wise exiting for classification tasks in large langua

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Model Releases

UniT: Toward a Unified Physical Language for Human-to-Humanoid Policy Learning and World Modeling

DGX agent

arXiv:2604.19734v1 Announce Type: cross Abstract: Scaling humanoid foundation models is bottlenecked by the scarcity of robotic data. While massive egocentric human data offers a scalable alternative,

model-releasesarxiv-cs-ai

22 Apr 2026

Unlocking the Edge deployment and ondevice acceleration of multi-LoRA enabled one-for-all foundational LLM

arXiv:2604.18655v1 Announce Type: cross Abstract: Deploying large language models (LLMs) on smartphones poses significant engineering challenges due to stringent constraints on memory, latency, and ru

Unveiling Fine-Grained Visual Traces: Evaluating Multimodal Interleaved Reasoning Chains in Multimodal STEM Tasks

arXiv:2604.19697v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) have shown promising reasoning abilities, yet evaluating their performance in specialized domains remains chall

URoPE: Universal Relative Position Embedding across Geometric Spaces

arXiv:2604.18747v1 Announce Type: new Abstract: Relative position embedding has become a standard mechanism for encoding positional information in Transformers. However, existing formulations are typi

VCE: A zero-cost hallucination mitigation method of LVLMs via visual contrastive editing

arXiv:2604.19412v1 Announce Type: cross Abstract: Large vision-language models (LVLMs) frequently suffer from Object Hallucination (OH), wherein they generate descriptions containing objects that are

VDPP: Video Depth Post-Processing for Speed and Scalability

arXiv:2604.06665v2 Announce Type: replace Abstract: Video depth estimation is essential for providing 3D scene structure in applications ranging from autonomous driving to mixed reality. Current end-t

VecHeart: Holistic Four-Chamber Cardiac Anatomy Modeling via Hybrid VecSets

arXiv:2604.19403v1 Announce Type: new Abstract: Accurate cardiac anatomy modeling requires the model to be able to handle intricate interrelations among structures. In this paper, we propose VecHeart,

VideoAgent: Personalized Synthesis of Scientific Videos

arXiv:2509.11253v2 Announce Type: replace Abstract: The technical complexity of research papers often limits their reach, necessitating more accessible formats like scientific videos to disseminate ke

ViDoRe V3: A Comprehensive Evaluation of Retrieval Augmented Generation in Complex Real-World Scenarios

arXiv:2601.08620v2 Announce Type: replace Abstract: Retrieval-Augmented Generation (RAG) pipelines must address challenges beyond simple single-document retrieval, such as interpreting visual elements

Visual Reasoning Agent: Robust Vision Systems in Remote Sensing via Inference-Time Scaling

arXiv:2509.16343v2 Announce Type: replace-cross Abstract: Building robust vision systems for high-stakes domains such as remote sensing requires stronger visual reasoning than what single-pass inferen

Visual-TableQA: Open-Domain Benchmark for Reasoning over Table Images

arXiv:2509.07966v2 Announce Type: replace-cross Abstract: Visual reasoning over structured data such as tables is a critical capability for modern vision-language models (VLMs), yet current benchmarks

VLA Foundry: A Unified Framework for Training Vision-Language-Action Models

arXiv:2604.19728v1 Announce Type: cross Abstract: We present VLA Foundry, an open-source framework that unifies LLM, VLM, and VLA training in a single codebase. Most open-source VLA efforts specialize

VLM Performance：Qwen3.6-27B is natively multimodal, supporting both vision-language thinking and non-thinking modes in a single unified chec…

VLM Performance：Qwen3.6-27B is natively multimodal, supporting both vision-language thinking and non-thinking modes in a single unified checkpoint — the same as Qwen3.6-35B-A3B. It handles images and

model-releasesqwen--x

Voice of India: A Large-Scale Benchmark for Real-World Speech Recognition in India

arXiv:2604.19151v1 Announce Type: new Abstract: Existing Indic ASR benchmarks often use scripted, clean speech and leaderboard driven evaluation that encourages dataset specific overfitting. In additi

Watch the Weights: Unsupervised monitoring and control of fine-tuned LLMs

arXiv:2508.00161v3 Announce Type: replace-cross Abstract: The releases of powerful open-weight large language models (LLMs) are often not accompanied by access to their full training data. Existing in

We just hit #1 on the @huggingface BrowseComp-Plus leaderboard. Best accuracy: 92.53%. Best recall: 88.79%. Lowest calibration error across …

model-releasesai21-labs--x

We just hit #1 on the @huggingface BrowseComp-Plus leaderboard. Best accuracy: 92.53%. Best recall: 88.79%. Lowest calibration error across all submissions. Built with @AI21Labs Maestro. https://huggi

We need open traces so that everyone can train open agent models! cc @steipete @badlogicgames @thdxr @matanSF @hwchase17

model-releasesclem-delangue--x

We need open traces so that everyone can train open agent models! cc @steipete @badlogicgames @thdxr @matanSF @hwchase17 People are misreading the SpaceX/Cursor deal as an M&A story. It’s actually a b

Welcome to the agentic BI era with Looker

By combining the analytical depth of Looker with Google’s Agentic Data Cloud, the potential to transform how we model, interact with, and act on our data appears limitless. This week at Google Cloud N

We've published new research on how we post-train models for accurate search-augmented answers. Our SFT + RL pipeline improves search, citat…

model-releasesperplexity--x

We've published new research on how we post-train models for accurate search-augmented answers. Our SFT + RL pipeline improves search, citation quality, instruction following, and efficiency. With Qwe

What’s new in BigQuery: Powering the Agentic Era

Succeeding in the agentic era requires a transformation in your data strategy: moving from human-scale to agent-first workloads, evolving from reactive intelligence to proactive action, and shifting f

What’s new in Cloud Run at Next ‘26

From vibe-coded and large-scale apps to AI models and agents, Cloud Run delivers on-demand compute with zero overhead and pay-per-use pricing for all of your workloads. Last year, the number of extern

What’s new in GKE at Next ‘26

This week at Google Cloud Next ‘26, we are sharing the evolution of Google Kubernetes Engine (GKE), delivering leading performance, efficiency, security, and scale for your most demanding and complex

What’s new in the Agentic Data Cloud: Powering the System of Action

Companies are shifting from gen AI that simply answers questions to autonomous agents that perceive, reason, and act on their behalf. Attempting to scale these agents on legacy stacks exposes structur

What’s new with compute: Scaling core and agentic workloads

At Google Cloud Next, we’re announcing a range of compute capabilities to enable your core general purpose and AI workloads for the agentic world with higher performance and lower costs. Why it matter

What’s new with the Cross-Cloud Network at Next ‘26

While generative AI sparked a revolution, the true paradigm shift is the rapid evolution from standalone AI models to multi-agent autonomous systems. In this new era, the network transcends basic conn

What’s next in Google AI infrastructure: Scaling for the agentic era

AI is evolving from answering questions to reasoning and taking action. Companies who want to lead in today’s agentic era require computing infrastructure designed and optimized for these new requirem

When and What to Ask: AskBench and Rubric-Guided RLVR for LLM Clarification

arXiv:2602.11199v2 Announce Type: replace Abstract: Large language models (LLMs) often respond even when prompts omit critical details or include misleading information, leading to hallucinations or r

When Safety Fails Before the Answer: Benchmarking Harmful Behavior Detection in Reasoning Chains

arXiv:2604.19001v1 Announce Type: new Abstract: Large reasoning models (LRMs) produce complex, multi-step reasoning traces, yet safety evaluation remains focused on final outputs, overlooking how harm

... which keeps things confusing, since it raises an important new question https://x.com/simonw/status/2046798283700617267

model-releasessimon-willison--x

... which keeps things confusing, since it raises an important new question https://x.com/simonw/status/2046798283700617267 @TheAmolAvasare If I sign up for a new $20/month account today and roll the

Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?

arXiv:2603.24472v2 Announce Type: replace Abstract: Self-distillation has emerged as an effective post-training paradigm for LLMs, often improving performance while shortening reasoning traces. Howeve

With Gemini Enterprise Agent Platform, Google brings agentic development and control under one roof

model-releasessiliconangle

Google Cloud is taking a massive leap toward building the autonomous enterprise with the launch of the Gemini Enterprise Agent Platform, an evolution of the existing Vertex AI platform that becomes it

Workspace agents can work across tools—pulling context from docs, email, chats, code, and systems, and taking approved actions like updating…

Workspace agents can work across tools—pulling context from docs, email, chats, code, and systems, and taking approved actions like updating @Linear issues, creating docs, or sending messages. In @Sla

model-releasesopenai--x

Wrote up Anthropic's self-own about Claude Code pricing from this afternoon on my blog - it turned out they'd reversed course just as I hit …

model-releasessimon-willison--x

Wrote up Anthropic's self-own about Claude Code pricing from this afternoon on my blog - it turned out they'd reversed course just as I hit publish, so I've tried to update it to reflect the current s

X launches Custom Timelines, a Grok-powered feature letting users pin any of over 75 topics to their home tab, in early access to Premium subscribers on iOS (Nikita Bier/@nikitabier)

Nikita Bier / @nikitabier: X launches Custom Timelines, a Grok-powered feature letting users pin any of over 75 topics to their home tab, in early access to Premium subscribers on iOS — Ladies and gen

model-releasestechmeme

Xiaomi's MiMo-V2.5 and MiMo-V2.5-Pro are both now available in Hermes Agent through Nous Portal and OpenRouter! Just `hermes update`!

model-releasesnous-research--x

Xiaomi's MiMo-V2.5 and MiMo-V2.5-Pro are both now available in Hermes Agent through Nous Portal and OpenRouter! Just `hermes update`! Xiaomi MiMo-V2.5 Series: Pushing Open-Source Agents Forward 🔸 MiMo

Xpertbench: Expert Level Tasks with Rubrics-Based Evaluation

arXiv:2604.02368v4 Announce Type: replace Abstract: As Large Language Models (LLMs) exhibit plateauing performance on conventional benchmarks, a pivotal challenge persists: evaluating their proficienc

ZC-Swish: Stabilizing Deep BN-Free Networks for Edge and Micro-Batch Applications

arXiv:2604.19453v1 Announce Type: new Abstract: Batch Normalization (BN) is a cornerstone of deep learning, yet it fundamentally breaks down in micro-batch regimes (e.g., 3D medical imaging) and non-I

A Benchmark Study of Segmentation Models and Adaptation Strategies for Landslide Detection from Satellite Imagery

arXiv:2604.16663v1 Announce Type: new Abstract: Landslide detection from high resolution satellite imagery is a critical task for disaster response and risk assessment, yet the relative effectiveness

A Comparative Evaluation of Geometric Accuracy in NeRF and Gaussian Splatting

arXiv:2604.18205v1 Announce Type: new Abstract: Recent advances in neural rendering have introduced numerous 3D scene representations. Although standard computer vision metrics evaluate the visual qua

A Note on TurboQuant and the Earlier DRIVE/EDEN Line of Work

arXiv:2604.18555v1 Announce Type: new Abstract: This note clarifies the relationship between the recent TurboQuant work and the earlier DRIVE (NeurIPS 2021) and EDEN (ICML 2022) schemes. DRIVE is a 1-

A Probabilistic Consensus-Driven Approach for Robust Counterfactual Explanations

arXiv:2604.17494v1 Announce Type: new Abstract: Counterfactual explanations (CFEs) are essential for interpreting black-box models, yet they often become invalid when models are slightly changed. Exis

A Real-World Grasping-in-Clutter Performance Evaluation Benchmark for Robotic Food Waste Sorting

model-releasesarxiv-cs-ro

arXiv:2602.18835v2 Announce Type: replace Abstract: Food waste management is critical for sustainability, yet inorganic contaminants hinder recycling potential. Robotic automation accelerates sorting

A Survey of Spatial Memory Representations for Efficient Robot Navigation

arXiv:2604.16482v1 Announce Type: new Abstract: As vision-based robots navigate larger environments, their spatial memory grows without bound, eventually exhausting computational resources, particular

A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the Foundation Model Era

arXiv:2604.16586v1 Announce Type: new Abstract: Molecular property prediction integrates quantum chemistry, cheminformatics, and deep learning to connect molecular structure with physicochemical and b

A Transformer and Prototype-based Interpretable Model for Contextual Sarcasm Detection

arXiv:2503.11838v2 Announce Type: replace Abstract: Sarcasm detection, with its figurative nature, poses unique challenges for affective systems designed to perform sentiment analysis. While these sys

Abstain-R1: Calibrated Abstention and Post-Refusal Clarification via Verifiable RL

arXiv:2604.17073v1 Announce Type: new Abstract: Reinforcement fine-tuning improves the reasoning ability of large language models, but it can also encourage them to answer unanswerable queries by gues

Adaptive Forensic Feature Refinement via Intrinsic Importance Perception

arXiv:2604.16879v1 Announce Type: new Abstract: With the rapid development of generative models and multimodal content editing technologies, the key challenge faced by synthetic image detection (SID)