model releases

The top announcements for startups from Google I/O ‘26

model-releasesgoogle-cloud-ai

Many of the world’s fastest-growing AI startups are choosing to build their future — and the world’s — on Google Cloud because of our complete and open AI stack. We embed AI into every layer of our ar

Human

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Model Releases

The Visual Iconicity Challenge: Evaluating Vision-Language Models on Sign Language Form-Meaning Mapping

DGX agent

arXiv:2510.08482v3 Announce Type: replace-cross Abstract: Iconicity, the resemblance between linguistic form and meaning, is pervasive in signed languages, offering a natural testbed for visual ground

model-releasesarxiv-cs-cl

21 May 2026

The Yes-Man Syndrome: Benchmarking Abstention in Embodied Robotic Agents

arXiv:2605.20544v1 Announce Type: cross Abstract: Vision-language models (VLMs) are used as high-level planners for embodied agents, translating natural language instructions and visual observations i

Theoretical guidelines for annealed Langevin dynamics in compositional simulation-based inference

arXiv:2605.21253v1 Announce Type: cross Abstract: Compositional score-based approaches to simulation-based inference (SBI) approximate the posterior over a shared parameter given n independent observa

THEval. Evaluation Framework for Talking Head Video Generation

arXiv:2511.04520v4 Announce Type: replace Abstract: Video generation has achieved remarkable progress, with generated videos increasingly resembling real ones. However, the rapid advance in generation

Time-Dependent PDE-Constrained Optimization via Weak-Form Latent Dynamics

arXiv:2605.20639v1 Announce Type: cross Abstract: Optimization problems constrained by high-dimensional, time-dependent partial differential equations require repeated forward and sensitivity solves,

TimeSRL: Generalizable Time-Series Behavioral Modeling via Semantic RL-Tuned LLMs -- A Case Study in Mental Health

arXiv:2605.21295v1 Announce Type: new Abstract: Longitudinal passive sensing enables continuous health prediction, yet models often fail under cross-dataset distribution shifts. Traditional ML overfit

Tippett-minimum Fusion of Representation-space Diffusion Models for Multi-Encoder Out-of-Distribution Detection

arXiv:2605.20502v1 Announce Type: cross Abstract: We address out-of-distribution (OOD) detection across the full spectrum of distribution shifts -- global domain changes, semantic divergence, texture

TLMs: Tiny LLMs and Agents on Edge Devices with @cormacb https://www.youtube.com/watch?v=-TiET_K-E_g Function Gemma ships at 270 million par…

TLMs: Tiny LLMs and Agents on Edge Devices with @cormacb https://www.youtube.com/watch?v=-TiET_K-E_g Function Gemma ships at 270 million parameters and runs nearly 2,000 tokens per second prefill on a

model-releasesswyx--x

To Select or not to Select, that is the Question: Distilling Robot Skill Prediction into a Small Ensemble

model-releasesarxiv-cs-ro

arXiv:2605.21242v1 Announce Type: new Abstract: As robot fleets become more heterogeneous, including humanoids, rovers, quadrupeds, and drones, selecting the right robot for a task becomes a core syst

Today we release a study on decoupling the benefits of subword tokenization for language model training, by simulating each suspected benefi…

model-releasesnous-research--x

Today we release a study on decoupling the benefits of subword tokenization for language model training, by simulating each suspected benefit one at a time inside a 1.7B byte-level pretraining pipelin

Token Maxxing, April 2026-May 2026, RIP 🪦

model-releasesgary-marcus--x

Token Maxxing, April 2026-May 2026, RIP 🪦 🦔Microsoft canceled its internal Claude Code licenses this week after token-based billing made the cost untenable, even for a company with effectively infinit

Towards UAV Detection in the Real World: A New Multispectral Dataset UAVNet-MS and a New Method

arXiv:2605.20963v1 Announce Type: new Abstract: The proliferation of unmanned aerial vehicles (UAVs) has created urgent demand for precise UAV monitoring. Existing RGB-based systems rely on spatial cu

Toxic Subword Pruning for Dialogue Response Generation on Large Language Models

arXiv:2410.04155v2 Announce Type: replace Abstract: How to defend large language models (LLMs) from generating toxic content is an important research area. Yet, most research focused on various model

Training distribution determines the ceiling of drug-blind cancer sensitivity prediction

arXiv:2605.20885v1 Announce Type: new Abstract: Precision oncology requires predicting which drugs will suppress a specific tumor from its molecular profile, but drug-blind sensitivity prediction has

Training Language Agents to Learn from Experience

arXiv:2605.20477v1 Announce Type: cross Abstract: Language agents can adapt from experience in interactive environments, but current reflection-based methods can only self-correct within a single task

TRAM: Test-Time Risk Adaptation with Mixture of Agents

arXiv:2408.08812v2 Announce Type: replace Abstract: Deployed reinforcement learning agents often face safety requirements that are specified only after training, such as new hazard maps, revised risk

Try Composer 2.5

model-releaseselon-musk--x

Try Composer 2.5 New CursorBench results just dropped. Two big takeaways. Composer 2.5 is way better than most people think. 63.2% score at $0.55 per task. Nearly matching Opus 4.7 Max and GPT 5.5 Ext

Tunable MAGMAX: Preference-Aware Model Merging for Continual Learning

arXiv:2605.20803v1 Announce Type: new Abstract: Continual learning (CL) aims to train models sequentially on multiple tasks while mitigating catastrophic forgetting of previously learned knowledge. Re

Under Pressure: Emotional Framing Induces Measurable Behavioral Shifts and Structured Internal Geometry in Small Language Models

arXiv:2605.20202v1 Announce Type: new Abstract: I study whether emotionally framed evaluation follow-ups change both the behavior and the calm-relative internal representations of small, locally deplo

Understanding and Improving Communication Performance in Multi-node LLM Inference

arXiv:2511.09557v4 Announce Type: replace-cross Abstract: As large language models (LLMs) continue to grow in size, distributed inference has become increasingly important. Model-parallel strategies m

Universal Reasoner: A Single, Composable Plug-and-Play Reasoner for Frozen LLMs

arXiv:2505.19075v3 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have demonstrated remarkable general capabilities, but enhancing skills such as reasoning often demands substanti

VDFP: Video Deflickering with Flicker-banding Priors

arXiv:2605.21079v1 Announce Type: new Abstract: Capturing digital screens with smartphones frequently induces severe banding due to hardware synchronization mismatches. Existing video restoration meth

Verifiable Provenance and Watermarking for Generative AI: An Evidentiary Framework for International Operational Law and Domestic Courts

arXiv:2605.21002v1 Announce Type: cross Abstract: Generative artificial intelligence now synthesizes photorealistic imagery, audio, and video at a cost that defeats traditional forensic intuition. The

VersusQ: Pairwise Margin Reasoning for Generalizable Video Quality Assessment

arXiv:2605.21130v1 Announce Type: new Abstract: Large Multimodal Models (LMMs) have shown promise for video quality assessment, but most methods still predict an absolute score for each video. Such po

Vision Transformers and Convolutional Neural Networks for Land Use Scene Classification

arXiv:2605.21268v1 Announce Type: new Abstract: Land Use Scene Classification (LUSC) from remote sensing imagery plays a critical role in environmental monitoring, urban planning, and sustainable reso

VISTA: Technical Report for the Ego4D Short-Term Object Interaction Anticipation at EgoVis 2026

arXiv:2605.20901v1 Announce Type: new Abstract: We propose VISTA, a V-JEPA Integrated StillFast Temporal Anticipator for the Ego4D Short-Term Object Interaction Anticipation (STA) Challenge at EgoVis

VISTAQA: Benchmarking Joint Visual Question Answering and Pixel-Level Evidence

arXiv:2605.20676v1 Announce Type: new Abstract: Establishing a clear link between model predictions and the visual evidence that supports them is critical for transparency and reliability in multimoda

VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models

model-releasesarxiv-cs-ro

arXiv:2605.20774v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models have shown strong promise for general-purpose robotic manipulation, but their real-world evaluation remains limited

VSCD: Video-based Scene Change Detection in Unaligned Scenes

arXiv:2605.20821v1 Announce Type: new Abstract: Detecting what has changed in an environment is essential for long-term autonomy, yet most change detection settings assume fixed viewpoints, mild misal

WaveGraphNet: Physics-Consistent Guided-Wave Damage Localization through Coupled Inverse-Forward Graph Learning

arXiv:2605.20311v1 Announce Type: new Abstract: Guided-wave structural health monitoring enables damage localization in composite plates using sparse networks of bonded piezoelectric transducers. Howe

WCXB: A Multi-Type Web Content Extraction Benchmark

arXiv:2605.21097v1 Announce Type: new Abstract: Web content extraction - isolating a page's main content from surrounding boilerplate - is a prerequisite for search indexing, retrieval-augmented gener

Weight Decay Regimes in Grokking Transformers: Cheap Online Diagnostics

arXiv:2605.20441v1 Announce Type: new Abstract: Transformers trained on modular arithmetic exhibit sharp transitions between memorization, generalization, and collapse. We show that weight decay acts

We’re launching the Google DeepMind Accelerator program in Asia Pacific to tackle environmental risks

model-releasesgoogle-deepmind

Google DeepMind has launched 'AI for the Planet,' a three-month accelerator program in Asia Pacific focused on leveraging advanced AI to combat environmental challenges like climate change, biodiversi

What Do Biomedical NER and Entity Linking Benchmarks Measure? A Corpus-Centric Diagnostic Framework

arXiv:2605.20537v1 Announce Type: new Abstract: Biomedical named entity recognition (NER) and entity linking (EL) strongly depend on annotated corpora, but the utility of these resources for benchmark

What Twelve LLM Agent Benchmark Papers Disclose About Themselves: A Pilot Audit and an Open Scoring Schema

arXiv:2605.21404v1 Announce Type: new Abstract: We read twelve well-known LLM agent benchmark papers and recorded, dimension by dimension, what each paper actually says about how its evaluation was ru

When Irregularity Helps: A Subclass Analysis of Inductive Bias in Neural Morphology

arXiv:2605.20558v1 Announce Type: new Abstract: Neural morphological generation systems often achieve high aggregate accuracy on benchmark datasets, yet such performance can conceal systematic errors

When to Retrain after Drift: A Data-Only Test of Post-Drift Data Size Sufficiency

arXiv:2603.09024v2 Announce Type: replace Abstract: Sudden concept drift makes previously trained predictors unreliable, yet deciding when to retrain and what post-drift data size is sufficient is rar

WikiVQABench: A Knowledge-Grounded Visual Question Answering Benchmark from Wikipedia and Wikidata

arXiv:2605.21479v1 Announce Type: new Abstract: Visual Question Answering (VQA) benchmarks have largely emphasized perception-based tasks that can be solved from visual content alone. In contrast, man

WildRoadBench: A Wild Aerial Road-Damage Grounding Benchmark for Vision-Language Models and Autonomous Agents

arXiv:2605.20306v1 Announce Type: new Abstract: We introduce WildRoadBench, a wild aerial road-damage grounding benchmark that couples direct visual grounding by vision-language models with autonomous

Winfree Oscillatory Neural Network

arXiv:2605.20922v1 Announce Type: cross Abstract: Oscillations and synchronization are widely believed to play a fundamental role in representation and computation. However, existing machine learning

Wordle 1,796 4/6 ⬛⬛⬛🟨🟨 ⬛⬛⬛⬛⬛ ⬛⬛🟩⬛🟨 🟩🟩🟩🟩🟩

model-releasesanthropic--x

This post shows a Wordle game result where the player solved puzzle #1,796 in 4 attempts, with the final answer being a five-letter word with the pattern shown in green squares. The emoji grid display

You Only Need Minimal RLVR Training: Extrapolating LLMs via Rank-1 Trajectories

arXiv:2605.21468v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has become a dominant paradigm for improving reasoning in large language models (LLMs), yet the

ZEBRA: Zero-shot Budgeted Resource Allocation for LLM Orchestration

arXiv:2605.20485v1 Announce Type: new Abstract: As autonomous agents increasingly execute end-to-end tasks under fixed monetary budgets, the pressing open question shifts from whether the budget is re

100 things we announced at I/O 2026

Google I/O 2026 unveiled new models, agents and tools to help users build, search, create, discover, shop and get more done. Key announcements included Gemini Omni, Google Antigravity, and Universal C

model-releasesgoogle-ai

A Bitter Lesson for Data Filtering

model-releasesarxiv-cs-ai

arXiv:2605.19407v1 Announce Type: cross Abstract: We investigate data filtering for large model pretraining via new scaling studies that target the high compute, data-scarce regime. In spite of an app

A Case for Agentic Tuning: From Documentation to Action in PostgreSQL

model-releasesarxiv-cs-ai

arXiv:2605.19988v1 Announce Type: cross Abstract: Documentation has long guided computer system tuning by distilling expert knowledge into per-parameter recommendations. Yet such guides capture only w

A Family of Divergence Measures for Evaluating the Reconstruction Quality of Explainable Ensemble Trees

arXiv:2605.19618v1 Announce Type: new Abstract: Validating interpretable surrogate models for ensemble learners requires measuring agreement between the ensemble's internal representation and its surr