model releases

Training-free Task Classification for Multi-Task Model Merging

arXiv:2606.22589v1 Announce Type: new Abstract: Ever since the advent of foundation models and the pre-training-finetuning paradigm, there have been numerous efforts to merge multiple task-specific ex

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Model Releases

Translating Inference-Time Control to Radiology Vision-Language Models: Activation Steering for Pneumonia Classification on Chest X-rays

DGX agent

arXiv:2606.20852v1 Announce Type: new Abstract: Inference-time engineering can alter model behavior without fine-tuning. However, its utility for improving diagnostic performance in medical vision-lan

model-releasesarxiv-cs-cv

23 Jun 2026

Two-Bridge: Exclusive Objectives and Extended Horizon StarCraft II Benchmark

arXiv:2603.06608v2 Announce Type: replace-cross Abstract: The research community lacks a middle ground between StarCraft II full game and its mini-games. The full-game's sprawling state-action space r

UBP2: Uncertainty-Balanced Preference Planning for Efficient Preference-based Reinforcement Learning

arXiv:2606.19328v2 Announce Type: replace Abstract: Preference-based RL provides an approach to learning reward models from pairwise comparisons of behaviors, bypassing the need for explicit reward de

Ultra-Fusion: A Resilient Tightly-Coupled Multi-Sensor Fusion SLAM Framework under Sensor Degradation and Spatiotemporal Perturbation for Intelligent Transportation Systems

model-releasesarxiv-cs-ro

arXiv:2606.21223v1 Announce Type: new Abstract: Reliable localization is essential for intelligent transportation systems (ITS), including autonomous vehicles, quadruped last-mile carriers, and infras

Understanding Parallel Samplers in Masked Diffusion via Random Walks on Graphs

arXiv:2606.22976v1 Announce Type: new Abstract: In this paper, we propose using random walks on graphs as a verifiable sandbox to study different parallel sampling strategies in masked diffusion model

UniRank: Unified Rank Allocation for Low-Rank LLM Compression

arXiv:2606.21847v1 Announce Type: new Abstract: Low-rank decomposition serves as a promising compression paradigm for large language models, however, rank allocation remains challenging: manual rules

UnityShots: Memory-Driven Multi-Shot Audio-Video Generation with Boundary-Aware Gating

arXiv:2606.21661v1 Announce Type: new Abstract: Generating a coherent multi-shot video requires structured cross-shot memory. Subject appearance, scene context, and speaker identity must persist acros

Unlimited OCR Works

arXiv:2606.23050v1 Announce Type: new Abstract: Recently, end-to-end OCR models, exemplified by DeepSeek OCR, have once again thrust OCR into the spotlight. A widely held view is that employing a larg

Vera: A Layered Diffusion Model for Content-Preserving Video Editing

arXiv:2606.23610v1 Announce Type: new Abstract: Video diffusion models have enabled remarkable progress in video generation and editing. However, content preservation remains a core challenge: existin

VeriEvol: Scaling Multimodal Mathematical Reasoning via Verifiable Evol-Instruct

arXiv:2606.23543v1 Announce Type: cross Abstract: Scaling reinforcement learning for visual mathematical reasoning requires more than generating harder questions: as data volume grows, the reward labe

Verifiable, private AI: Google Cloud expands Confidential Computing frontiers

model-releasesgoogle-cloud-ai

Protecting sensitive data used with AI is a critical part of our commitment to providing advanced and secure cloud infrastructure. Confidential Computing cryptographically protects data in use in hard

VideoAgent: All-in-One Framework for Video Understanding and Editing

arXiv:2606.23327v1 Announce Type: new Abstract: Video editing has become essential in digital media creation, yet existing automated systems are restricted to short segment processing and domain-speci

Vision-language models for chest radiography do not always need the image

arXiv:2606.17710v2 Announce Type: replace Abstract: Medical vision-language models report strong chest radiograph accuracy, and this is increasingly read as evidence that they use the image. That infe

VolHuMe: a High-Resolution Large Scale Dataset of Volumetric Human Meshes

arXiv:2606.23062v1 Announce Type: cross Abstract: We introduce VolHuMe, a dataset of high-quality 4D human scans captured with a state-of-the-art volumetric studio using 64 RGB and 32 depth cameras. V

Weighted Score-Oriented Losses for Temporally Localized Event Prediction

arXiv:2606.23145v1 Announce Type: new Abstract: Operational event-detection systems are rarely assessed by pointwise accuracy alone. In anomaly detection, changepoint detection, and warning systems, t

We're launching Claude Tag today. Tag Claude into Slack and it works in channel with you. It’s proactive, multiplayer, with its own identity…

model-releasesboris-cherny--x

We're launching Claude Tag today. Tag Claude into Slack and it works in channel with you. It’s proactive, multiplayer, with its own identity and memory. But it’s not just a bot in Slack. Over the last

We’ve worked hard to make it secure at every level. 1/ At the model training stage, 2/ the classifiers on top of our models and things like …

model-releasesboris-cherny--x

We’ve worked hard to make it secure at every level. 1/ At the model training stage, 2/ the classifiers on top of our models and things like auto mode, 3/ we protect what Claude has access to (websites

When AUC 0.998 Is Not Enough: A Candidate Evaluation Protocol for Hidden-State Probes of Indirect Prompt Injection in Multimodal Computer-Use Agents

arXiv:2606.22864v1 Announce Type: new Abstract: Hidden-state probing -- a linear classifier on a frozen vision-language model's internal activations -- has emerged as an attractive evaluation tool for

When Robots Rate Their Own Interactions: Engagement Validity and the Strangeness Failure

model-releasesarxiv-cs-ro

arXiv:2606.23339v1 Announce Type: new Abstract: Human-robot interaction (HRI) evaluation relies almost exclusively on human-completed questionnaires, leaving the robot's perspective unexamined. We pro

When Web Agents Finish but Still Fail: Reproducible Triggers and Trace Diagnostics for Parallel Web Exploration

arXiv:2606.20724v1 Announce Type: cross Abstract: Long-horizon web agents often fail in ways hidden by final-answer evaluation: they may visit useful pages, produce a well-formed answer, and terminate

Where Does the Signal Live? A Web Data Recipe for Medical Encoder Pretraining

arXiv:2606.22079v1 Announce Type: cross Abstract: Web data curation has been widely studied for decoder Large Language Model (LLM) pretraining. Encoders for dense-terminology domains such as medicine,

Who Owns the AI Recommendation? A Multi-Industry Empirical Map of Brand Category Ownership Across Large Language Models

arXiv:2606.23057v1 Announce Type: cross Abstract: Large language models now mediate how buyers discover products and services, making the competitive structure of AI-generated recommendations a strate

Why the structure matters: OCR 4 localizes each block with a bounding box, classifies it (title, table, equation, signature…), and scores co…

model-releasesmistral-ai--x

Why the structure matters: OCR 4 localizes each block with a bounding box, classifies it (title, table, equation, signature…), and scores confidence per region, the foundation for source-grounded cita

WildBox: A Dataset and Benchmark for Aerial Monocular 3D Detection of African Savanna Wildlife

arXiv:2606.21309v1 Announce Type: new Abstract: We introduce WildBox, a dataset and benchmark for monocular 3D detection of wildlife from drone video, comprising 237,505 3D bounding box annotations ac

With agentic coding, complexity compounds in a mechanical way: unnecessary code ends up in the codebase, moves to the context window, degrad…

model-releasesfrancois-chollet--x

With agentic coding, complexity compounds in a mechanical way: unnecessary code ends up in the codebase, moves to the context window, degrades the model's reasoning abilities, leads to more unnecessar

Wordle 1,829 3/6 ⬛🟨🟩⬛🟩 ⬛⬛⬛⬛🟨 🟩🟩🟩🟩🟩

model-releasesanthropic--x

This post documents a Wordle game result where the player solved puzzle #1,829 in 3 attempts, with the final answer being a five-letter word shown in green squares. The notation uses the standard Word

WorkBenchMark: A LEGO-Based Assembly Benchmark with an Assembly-by-Disassembly Baseline for the Smart Manufacturing League

model-releasesarxiv-cs-ro

arXiv:2606.19358v2 Announce Type: replace Abstract: We introduceWorkBenchMark, a LEGO Duplo-based robotic assembly benchmark motivated by the RoboCup Smart Manufacturing League. Robotic assembly coupl

You can also tune in from home. The opening keynote will be livestreamed on September 29.

OpenAI announced that their opening keynote scheduled for September 29 will be livestreamed, allowing remote viewers to participate from home. This accessibility option expands attendance beyond in-pe

model-releasesopenai--x

Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer

arXiv:2511.22699v4 Announce Type: replace Abstract: The landscape of high-performance image generation models is currently dominated by proprietary systems, such as Nano Banana Pro and Seedream 4.0. L

Zero-order Parameter-free Optimization for LMO-based Methods: Novel Approach for Efficient Fine-tuning

arXiv:2606.14970v2 Announce Type: replace Abstract: Fine-tuning large language models (LLMs) has become a central application of modern optimization, enabling pretrained models to adapt to diverse dow

Zero-Shot Vision-Language Models for Classroom Engagement Recognition: A Benchmark Study of Prompt Sensitivity and Cross-Dataset Generalization

arXiv:2606.21861v1 Announce Type: new Abstract: Automated classroom engagement recognition holds substantial promise for scalable learning analytics, yet the suitability of modern Vision-Language Mode

Ai2 just released TMax 27B on Hugging Face A 27B terminal agent that hits 42.7% on Terminal Bench 2.0, rivaling models 40× its size.

model-releasesclem-delangue--x

AI2 released TMax 27B, a 27 billion parameter terminal agent model available on Hugging Face that achieves 42.7% performance on Terminal Bench 2.0, matching the capabilities of much larger models desp

All these new models landing this year but Flux Klein 9b FP8 has spoiled me. All I care about now is whether a new model can edit and be used on an 8GB GPU.

model-releasesr-stablediffusion

This Reddit post discusses user preferences for AI image generation models in 2026, expressing that despite numerous new model releases, the Flux Klein 9b FP8 model has become their benchmark for what

Another new idea to push the state of AI architectures forward. Sakana released a model that effectively uses a mixture of models to get wor…

model-releasesdavid-ha--x

Another new idea to push the state of AI architectures forward. Sakana released a model that effectively uses a mixture of models to get work done. You get a single API but then the work gets farmed o

As AI becomes integrated into every industry, without a sovereign solution, you run the risk of your infrastructure shutting down at a momen…

As AI becomes integrated into every industry, without a sovereign solution, you run the risk of your infrastructure shutting down at a moment's notice. Cohere CEO @aidangomez live at @FII_Institute1:

model-releasescohere--x

• At this point SpaceX just looks like CoreWeave with a bigger budget and a satellite company thrown in. • If they were anywhere near AGI th…

model-releasesgary-marcus--x

• At this point SpaceX just looks like CoreWeave with a bigger budget and a satellite company thrown in. • If they were anywhere near AGI they wouldn’t be leasing out so much capacity. • So much for t

Benchmarks tell only part of the story. Fugu’s real value shows up in long, messy, real-world workflows. During our beta with 500 users, we …

model-releasesdavid-ha--x

Benchmarks tell only part of the story. Fugu’s real value shows up in long, messy, real-world workflows. During our beta with 500 users, we saw Fugu Ultra drive meaningful progress in fully automated

Boost BigQuery with Python: Managed Python UDFs now generally available

model-releasesgoogle-cloud-ai

SQL is the industry standard for high-performance structured data analysis. However, expressing complex procedural logic, scientific computations, advanced string manipulations, or machine learning wo

Daybreak: Tools for securing every organization in the world

Daybreak is an OpenAI initiative focused on developing and distributing security tools designed to protect organizations globally from cyber threats. The program aims to democratize access to advanced

model-releasesopenai

Embed the world: Multimodal AI for searchable aerial imagery at scale

model-releasesaws-ml-blog

In this post, we walk through the problem space, our architecture on Amazon Bedrock and Amazon OpenSearch Serverless, the evaluation methodology we built on OpenStreetMap ground truth, four experiment

GLM-5.2 has been the most popular new model on Fireworks this past week. @ArtificialAnlys confirms why: #3 overall on GDPval-AA (1524 Elo), …

model-releasesfireworks-ai--x

GLM-5.2 has been the most popular new model on Fireworks this past week. @ArtificialAnlys confirms why: #3 overall on GDPval-AA (1524 Elo), #1 open weights by 116 points. Interest is showing no signs

GLM-5.2 leads open weights models and sits at #3 overall on GDPval-AA, a real-world agentic work benchmark GLM-5.2 from @Zai_org scores 1524…

model-releasesclem-delangue--x

GLM-5.2 leads open weights models and sits at #3 overall on GDPval-AA, a real-world agentic work benchmark GLM-5.2 from @Zai_org scores 1524 Elo on GDPval-AA, which measures performance on real-world,

GLM is the kind of model that revives serious interest in open source AI. It passes the blind test relative to the frontier models on the me…

model-releasesclem-delangue--x

GLM is the kind of model that revives serious interest in open source AI. It passes the blind test relative to the frontier models on the median production grade knowledge worker task. It’s affordable

Gray Swan: Red-Teaming after Mythos & the coming AI security crisis https://www.latent.space/p/gray-swan @GraySwanAI cofounders @zicokolter …

Gray Swan: Red-Teaming after Mythos & the coming AI security crisis https://www.latent.space/p/gray-swan @GraySwanAI cofounders @zicokolter and Matt Fredrikson explain why AI security is fundamentally

model-releasesswyx--x

Here is the prompt method behind this AR try-on app. The trick is not a magic prompt. It is the architecture of the prompt, and it works acr…

model-releaseszhipu-ai--x

Here is the prompt method behind this AR try-on app. The trick is not a magic prompt. It is the architecture of the prompt, and it works across GLM-5.2 and other frontier models. Full prompt: http://c

I printed a custom t-shirt that's an ode to @simonw's Pelican benchmark. My partner says he doesn't get it. But y'all get it, right? RIGHT!?

model-releasessimon-willison--x

Pamela Fox created a custom t-shirt design referencing Simon Willison's Pelican benchmark, a technical tool or metric in web development or performance testing. She posted about it on X (formerly Twit

In a rare joint statement, Five Eyes leaders warn AI models capable of taking down governments and businesses are mere months away, urging leaders to 'act now' (Sarah Basford Canales/The Guardian)

Sarah Basford Canales / The Guardian: In a rare joint statement, Five Eyes leaders warn AI models capable of taking down governments and businesses are mere months away, urging leaders to “act now” —

model-releasestechmeme