Event History
Browse verified industry events and important research by month, area, or content type.
Filter by domain
Start with all, official, or research. Expand to filter by domain.
Introducing Cosmos 3 Edge
NVIDIA announced Cosmos 3 Edge on Hugging Face on 2026-07-20.
NEW · Updated Jul 20, 2026 · OpenAISafety and alignment in an era of long-horizon models
OpenAI published a blog post on July 20, 2026, sharing lessons from deploying long-running AI models, highlighting new safety risks, observed failures, and improved safeguards through iterative deployment.
NEW · Updated Jul 17, 2026 · CAV-STIXGenEvaluating Open-Weight LLMs for Generating Structured Threat Information for Autonomous Vehicle Vulnerabilities
A study evaluated 11 open-source large language models (4B to 120B parameters) for converting CVE vulnerability descriptions of connected autonomous vehicles (CAVs) into STIX structured threat information, constructed the CAV-STIXGen dataset, and reported F1 scores for single-model configurations.
NEW · Updated Jul 17, 2026 · MuonWhen Does Muon Help Agentic Reinforcement Learning?
An exploratory study on ALFWorld sparse-reward agentic reinforcement learning tasks, using the Qwen2.5-0.5B-Instruct model, compared the Muon optimizer with AdamW. Under Group-in-Group Policy Optimization (GiGPO), applying Muon only to hidden weight matrices improved final window validation success rate from 0.290 to 0.546 (+88%). The effect depends on advantage estimator and learning rate: at learning rate 3e-5, Muon improved GRPO success rate from 0.161 to 0.268; at learning rate 1e-5, GraphGPO Muon achieved 0.901, normalized validation AUC increased from 0.399 to 0.556, and reaching 0.5 and 0.75 success rates was 30 and 60 updates earlier, respectively.
NEW · Updated Jul 17, 2026 · GPT-5.5, Claude Fable 5An Exam for Active Observers
The paper 'ActiveVision' releases a benchmark containing 17 tasks designed to measure the active observation ability of multimodal large language models. Evaluation results show that GPT-5.5, at the highest reasoning effort level, solves only 10.6% of tasks and scores zero on 11 of the 17 tasks; Claude Fable 5 solves only 3.5%; while three human participants average 96.1%. Even when models write and run their own visual code, the gap persists.
NEW · Updated Jul 17, 2026 · arXivWhen Do Multi-Agent Systems Help? An Information Bottleneck Perspective
An arXiv paper analyzes the conditions under which multi-agent systems (MAS) outperform single-agent systems (SAS) from an information bottleneck perspective, noting that MAS can improve efficiency by compressing context under limited relay bandwidth, but risk losing task-relevant information. It validates this through 18 controlled experiments across 5 benchmarks and 3 model scales.
NEW · Updated Jul 17, 2026 · ToolSciVerToolSciVer: Multimodal Scientific Claim Verification with Visual Tool Augmented Reinforcement Learning
ToolSciVer proposes the first tool-augmented framework for multimodal scientific claim verification. Experiments on SciVer and MuSciClaims datasets with five VLMs from Qwen, InternVL, and Gemma model families achieve superior performance compared to four baseline methods.
NEW · Updated Jul 17, 2026 · arXivA Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governance
An arXiv paper proposes a methodology for auditable trustworthiness levels in AI governance, including a formal framework and a lightweight lifecycle governance procedure, using decision trees as a proof-of-concept model to generate interpretable trustworthiness platforms, level transitions, and two diagnostics: boundary margin and profile drift.
NEW · Updated Jul 17, 2026 · CRAFTCRAFT: Clustering Rubrics to Diagnose Weak LLM Capabilities and Generate Targeted Fine-Tuning Data
The CRAFT method transforms rubric-based evaluation datasets into model-specific weak capability diagnoses by extracting capability descriptions, clustering them into a hierarchical capability tree, scoring, and selecting low-scoring nodes to guide the generation of targeted supervised fine-tuning data. It was compared with prompt-level EvalTree clustering and random generation on four open-source models, two specialized domains (finance and law), and 13 held-out benchmarks.
NEW · Updated Jul 17, 2026 · arXivHarmonizing AI Safety Thresholds
An arXiv paper proposes a method to harmonize safety thresholds across frontier AI companies in three risk areas: cyber misuse, biological misuse, and automated AI R&D, addressing inconsistencies that hinder verification and comparison.
NEW · Updated Jul 17, 2026 · arXivUnderstanding Reasoning from Pretraining to Post-Training
A study uses chess as a controlled testbed to examine how pretraining choices affect reinforcement learning (RL) post-training for reasoning. Language models from 5M to 1B parameters are pretrained on human chess games, fine-tuned on synthetic reasoning traces, and then RL-trained on chess puzzles with verifiable rewards. The post-RL performance at a given RL compute level is well-predicted by pretraining loss, and the slope of RL reward curves improves with model scale.
NEW · Updated Jul 17, 2026 · DADiffDADiff: Diffusion-Driven Cross-Domain Policy Adaptation for Reinforcement Learning
A paper titled 'DADiff: Diffusion-Driven Cross-Domain Policy Adaptation for Reinforcement Learning' was published on arXiv on 2026-07-17. It introduces a diffusion-based framework for online dynamics adaptation in reinforcement learning, using generative trajectory discrepancy to estimate dynamics mismatch between source and target domains. The framework includes reward modification and data selection variants, and provides a theoretical bound on policy performance difference based on generative trajectory deviation.
NEW · Updated Jul 17, 2026 · HCIGHCIG: A Hierarchical Cross-Modal Incongruity Graph Network for Multimodal Sarcasm and Cyberbullying Detection
A paper titled 'HCIG: A Hierarchical Cross-Modal Incongruity Graph Network for Multimodal Sarcasm and Cyberbullying Detection' was published on arXiv on 2026-07-17. It proposes HCIG, a framework that models cross-modal incongruity at token, phrase, and global levels using graph attention networks, and GCCN, a complementary architecture using contradiction-aware pooling. The models are evaluated on the MMSD sarcasm benchmark and the MultiBully cyberbullying dataset, with ablation studies and cross-task transfer experiments.
NEW · Updated Jul 17, 2026 · NVIDIA, Hugging FaceFine-tune video and image models at scale with NVIDIA NeMo Automodel and 🤗 Diffusers
NVIDIA NeMo Automodel and Hugging Face's Diffusers can be used for large-scale fine-tuning of video and image models.
NEW · Updated Jul 17, 2026 · arXivLLM-Powered Agentic AI for 5G/6G Networks: A Tutorial and Survey on Architectures, Protocols, and Standardization
A two-part tutorial and survey on arXiv formalizes agentic AI for 5G/6G networks, covering control planes, agentic foundations, and standardization alignment.
NEW · Updated Jul 17, 2026 · arXivSpatial Normalization for Cross-Domain Retinal Layer Segmentation in Optical Coherence Tomography
A study investigates spatial normalization as a preprocessing strategy to mitigate geometric domain shifts in retinal layer segmentation from Optical Coherence Tomography (OCT). The proposed fovea-centered normalization framework aligns OCT volumes into a common anatomical reference. The research evaluates state-of-the-art deep learning architectures and combines conventional overlap-based metrics for comprehensive assessment.
NEW · Updated Jul 17, 2026 · Qwen3-8BWhen Model Merging Rivals Joint Multi-Task Reinforcement Learning: A Task-Vector Geometry Analysis
A study on the AppWorld agent benchmark using Qwen3-8B specialists trained with LOOP found that model merging (TIES, RAM+) achieves task-goal completion statistically indistinguishable from joint multi-task reinforcement learning. The specialists' task vectors are near-orthogonal (cosine 0.06–0.10) with ~65% support overlap, causing sign-based merging methods to approximate uniform averaging.
NEW · Updated Jul 17, 2026 · BusinessCaseBenchFrontier AI performance across the business disciplines: a case-grounded benchmark of knowledge work and analytical reasoning
A new benchmark, BusinessCaseBench, measures LLM performance on analytical knowledge work using business cases across 18 disciplines. Frontier AI models score highly against instructor rubrics.
NEW · Updated Jul 17, 2026 · SciForgeSciForge: An AI-Native, Multimodal Workbench for Scientific Discovery
SciForge is a multimodal research-native AI workbench that reserves the graphical interface for human judgment while search, parsing, model routing, workflow execution, plotting, writing, and presentation generation run as modular agent-accessible services. It is built around five pillars: goal-scoped scientific decision governance, translate-then-reason for multimodal input, evidence governance for auditable traceability, collaborative team science with multi-role decision governance, and real-world application. The paper was published on arXiv on 2026-07-17.
NEW · Updated Jul 17, 2026 · arXivRevisiting data-driven dynamic security assessment with a tabular foundation model
A single tabular foundation model (TFM) achieves an average Macro F1 score of about 90% for pre-fault dynamic security assessment on the IEEE 68-bus system using only 120 labelled samples per contingency, eliminating the need for separate models per contingency and enabling in-context learning without retraining.
NEW · Updated Jul 17, 2026 · QEWCRethinking Quantum Continual Learning with Quantum Fisher Information
A paper proposes quantum elastic weight consolidation (QEWC), a quantum Fisher information (QFI)-informed regularization method to mitigate catastrophic forgetting in variational quantum classifiers (VQCs) during sequential task training. Simulations on classical image and quantum phase classification tasks show QEWC and classical Fisher information-based EWC both improve retention over no regularization, with mechanistic differences.
NEW · Updated Jul 17, 2026 · OpenAIA scorecard for the AI age
OpenAI CFO Sarah Friar introduces a practical AI scorecard to measure ROI through useful work, cost per successful task, reliability, and compute return rate.
NEW · Updated Jul 17, 2026 · AgentFAIRAgentFAIR: A Multi-Agent Collaborative Framework for FAIRness Evaluation of Geospatial Datasets
AgentFAIR is a multi-agent framework for evaluating FAIR compliance of geospatial datasets. It uses structured metadata extraction and 13 sub-principle-specific LLM evaluators, each producing a 0-3 maturity score, cited evidence, and recommendations. A critic agent checks evidence and consistency, and can request re-evaluation. On 50 datasets from 10 repositories, mean Findability, Accessibility, Interoperability, and Reusability scores were 79.7%, 70.4%, 45.3%, and 72.0%. Rank correlations with four baseline tools ranged from 0.31 to 0.61; the FAIR-enough comparison was not statistically significant. On a 10-dataset repeated-run subset, sub-principle agreement averaged 89%.
NEW · Updated Jul 17, 2026 · MoD-VLLMModularized Dynamic-Granularity Video LLM for Multi-Event Long Video Understanding
A new framework called MoD-VLLM is proposed for multi-event long video understanding. It uses a modularized dynamic-granularity approach that unifies temporal grounding and semantic understanding iteratively and self-reflectively. The framework includes a Positive-Negative Video Segments Grounding module and a Modularized Dynamic-Granularity Reflection module, forming a closed loop to progressively localize question-related video segments.
NEW · Updated Jul 17, 2026 · NeurOWLNeurOWL: An LLM-Based Neural-symbolic Framework for Incomplete OWL Ontology Reasoning
A paper titled 'NeurOWL: An LLM-Based Neural-symbolic Framework for Incomplete OWL Ontology Reasoning' was published on arXiv on 2026-07-17. It proposes an end-to-end neuro-symbolic framework that jointly performs verification and abduction for incomplete OWL ontologies, leveraging Large Language Models and ontology embeddings. The framework addresses subsumption reasoning without requiring a predefined candidate set of missing axioms. Evaluation on real-world ontologies across multiple domains shows strong and robust performance.
NEW · Updated Jul 17, 2026 · AquaAugmentorAquaAugmentor: A Novel Feature Augmentation Algorithm for Water Potability Prediction
A paper introduces AquaAugmentor, a feature augmentation algorithm designed to improve machine learning and deep learning models for water potability classification on low-dimensional datasets. The study uses a dataset with chemical attributes (pH, hardness, solids, chloramines, sulfate, etc.) and evaluates models with and without AquaAugmentor based on test accuracy and AUC score.
NEW · Updated Jul 16, 2026 · SciDiagramEditGPT-5.6 enters Microsoft 365 Copilot: the agent gets enterprise-grade distribution
GPT-5.6 becomes the preferred model for Microsoft 365 Copilot, covering Word, Excel, PowerPoint, Chat and Cowork.
NEW · Updated Jul 16, 2026 · MetaAutoSynthesis: An agentic system for automated meta-analysis
AutoSynthesis is an end-to-end multi-agent system for automated meta-analysis. It formulates search strategy, retrieves literature, screens studies, extracts statistics, computes effect sizes, and performs random-effects meta-analysis. It screened over 28 studies and extracted more than 20 quantitative claims. Pooled effect estimates are similar to Hedges' g of expert-conducted meta-analyses.
NEW · Updated Jul 16, 2026 · OpenAIWhy Teens Deserve Access to Safe AI
OpenAI is developing safety protections, learning tools, parental controls and expert collaborations for ChatGPT suitable for teenagers.
NEW · Updated Jul 16, 2026 · Hugging FaceNewer Models, Same Advantage
Hugging Face released a blog post "Newer Models, Same Advantage"
NEW · Updated Jul 16, 2026 · Google DeepMind, Isomorphic LabsOur approach to bioresilience
On July 16, 2026, Google DeepMind and Isomorphic Labs jointly published a blog post outlining their shared approach to bioresilience and AI models.
NEW · Updated Jul 16, 2026 · Google DeepMindOur approach to bioresilience
Google DeepMind and Isomorphic Labs published a blog post titled 'Our approach to bioresilience' on July 16, 2026, sharing their joint approach to bioresilience and AI models.
NEW · Updated Jul 16, 2026 · Cars24, OpenAIHow Cars24 scales conversations and builds faster with OpenAI
Cars24 uses OpenAI's voice and chat agents to process 1 million minutes of conversations per month, recovering 12% of lost potential customers and bringing agent workflows to teams across the company.
NEW · Updated Jul 16, 2026 · Hugging FaceSecurity Incident Disclosure — July 2026
Hugging Face released the security incident disclosure for July 2026.
NEW · Updated Jul 15, 2026 · Hugging FaceWhat building Shippy taught us about building agents
Hugging Face published a blog post discussing the insights from the Shippy project on building agents.
NEW · Updated Jul 15, 2026 · IBM ResearchModel Routing Is Simple. Until It Isn’t.
IBM Research published an article on Hugging Face blog discussing that model routing performs well in simple scenarios but may fail under complex conditions.
NEW · Updated Jul 15, 2026 · OpenAIThe US is advancing AI safety through state and federal action
OpenAI published a blog post on 2026-07-15 outlining a 'reverse federalism' approach to AI governance, where state laws help build a national framework for safe, democratic AI.
NEW · Updated Jul 15, 2026 · OpenAIGPT-Red: Unlocking Self-Improvement for Robustness
OpenAI released GPT-Red on 2026-07-15, an automated red-teaming system using self-play to enhance AI safety, alignment, and prompt injection robustness.
NEW · Updated Jul 15, 2026 · Thinking MachinesWelcome Inkling by Thinking Machines
Thinking Machines announces Inkling on Hugging Face blog.
NEW · Updated Jul 15, 2026 · Hugging FaceIntroducing Real World VoiceEQ: Measuring the human quality of voice AI
Hugging Face published a blog post on 2026-07-15 introducing Real World VoiceEQ, a metric for measuring the human quality of voice AI.
NEW · Updated Jul 14, 2026 · Resist and Update: Counterfactual Report Coordinates for Incentive-Compatible LLMsResist and Update: Counterfactual Report Coordinates for Incentive-Compatible LLMs
Aligned language models misreport under non-evidential incentive pressure. A method called counterfactual report-coordinate (CRC) clamp is introduced to enforce incentive-compatibility by resisting forbidden influences and updating on genuine evidence. The method is evaluated on a Bayesian-witness benchmark.
NEW · Updated Jul 14, 2026 · FormalAnalyticGeoFormalAnalyticGeo: A Neural-Symbolic Based Framework for Multimodal Analytic Geometry Problem Generation
FormalAnalyticGeo is a neural-symbolic framework for automatic generation of multimodal analytic geometry problems. It uses CDL (Condition Description Language) as a formal intermediate representation and an SDF (Signed Distance Field) engine for diagram rendering. The framework includes four LLM components: Generator, Formalizer, Measurer, and Quality Verifier.
NEW · Updated Jul 14, 2026 · PythiaA Multi-Agent System for Autonomous, Fine-Tuning-Free Clinical Symptom Detection: Development and Validation Study
Pythia is a multi-agent system for autonomous, fine-tuning-free clinical symptom detection from clinical notes. It runs on a locally hosted open-weights model and selects prompts using development-set sensitivity and specificity. In a study with 72 signs and symptoms from 400 clinical notes (387 patients), Pythia achieved mean sensitivity 0.76 and specificity 0.95, compared to a lexicon's 0.82 and 0.76. For 14 concepts where the lexicon labeled every note positive, Pythia recovered mean specificity 0.97.
NEW · Updated Jul 14, 2026 · OpenAIHow to manage AI investments in the agentic era
OpenAI published an article on July 14, 2026, discussing how to manage AI investments in the agentic era, proposing to manage enterprise AI investments by measuring useful work per dollar, improving efficiency, and scaling high-value workflows.
NEW · Updated Jul 14, 2026 · OpenAIHow data science teams use ChatGPT Work
OpenAI released two Academy articles on 2026-07-14, introducing how data science teams and sales teams use ChatGPT Work to complete tasks such as root cause briefs, impact reports, KPI memos, pipeline briefs, and meeting preparation packages.
NEW · Updated Jul 13, 2026 · MM-ToolSandBoxMM-ToolSandBox: A Unified Framework for Evaluating Visual Tool-Calling Agents
MM-ToolSandBox is a benchmark and evaluation framework for visually grounded tool-calling agents. It provides a stateful execution environment spanning 500+ tools across 16 application domains, supporting multi-image, multi-turn tasks. An automated scenario generation pipeline produces 258 human-verified nominal scenarios and 50 variants. Evaluating 12 state-of-the-art models shows the best model achieves below 50% success rate. Failure analysis reveals 53% of failures stem from incorrect information extraction from images.
NEW · Updated Jul 13, 2026 · IAANEncoder-Side Neuron Identification and Amplification for Acoustic Perception in Large Audio-Language Models
IAAN (Identifying and Amplifying Acoustic Neurons) is a training-free, label-free method that scores feed-forward neurons in the audio encoder by contrasting activation on real waveform vs noise reference, then amplifies top-scoring neurons at inference. On ten non-semantic speech attributes, IAAN improves average accuracy by 25.7 points on Audio-Flamingo-3, 21.4 on Qwen2.5-Omni, and 9.7 on Kimi-Audio.
NEW · Updated Jul 13, 2026 · StoryTellerStoryTeller: Training-Free Narrative Grounding for Long-Form Audio Description
StoryTeller is a training-free framework for long-form audio description that maintains a verified narrative memory to carry story-relevant information across scenes, using only raw video and a movie title, with optional retrieval of public movie metadata. It requires no subtitles, scripts, AD transcripts, aligned captions, character banks, precomputed face identities, or task-specific fine-tuning.
NEW · Updated Jul 13, 2026 · VoxENES 2026VoxENES 2026: Benchmarking Generalization of Speech Spoofing Detectors Against LLM-Era TTS and Voice Conversion
VoxENES 2026 is a bilingual (English and Spanish) benchmark of 53,628 audio samples generated using 10 contemporary speech synthesis methods and evaluated under 10 standardized post-processing conditions. Eight pretrained detectors were benchmarked without fine-tuning; the best model achieved 28.98% EER overall, while most performed near or below random chance.
NEW · Updated Jul 13, 2026 · XiaomiXiaomi-Robotics-U0: Unified Embodied Synthesis with World Foundation Model
Xiaomi-Robotics-U0 is a 38-billion-parameter multimodal autoregressive model for unified embodied synthesis, jointly optimizing text-to-image generation, image editing, embodied scene generation, embodied transfer, and embodied video generation. It is the first model to support high-quality multi-view scene generation across multiple robot embodiments and introduces structured, controllable embodied transfer. It achieves state-of-the-art results.
NEW · Updated Jul 13, 2026 · MAGICMAGIC: Transition-Aware Generation of Navigable Multi-Scene Game Worlds with Large Language Models
MAGIC is a four-stage pipeline that turns a single natural-language prompt into a runnable multi-scene game project, addressing cross-scene consistency, in-scene navigability, and transition evaluation. It plans a shared transition-aware intermediate representation, specifies each scene while enforcing portal reachability with a flood-fill validator, generates scenes together with transition scripts, and produces a runnable project.
NEW · Updated Jul 13, 2026 · HCRMapHCRMap: Pressure-Aware Hot-Expert Residency Mapping for 3.5D MoE Chiplet Inference
HCRMap is a hot expert residency mapping framework for pressure-aware expert replica management in 3.5D MoE inference. It dynamically determines expert promotion, retention, demotion, or eviction based on expert hotness, weight loading cost, migration overhead, and runtime resource pressure. Experimental results show HCRMap reduces end-to-end latency by 43.6% and 43.0% over Hydra in prefill and decode stages.
NEW · Updated Jul 13, 2026 · Google DeepMindEmpowering India’s next generation of innovators with ATL Saathi
Google and AIM jointly launch ATL Saathi, a Gemini-based AI tool aimed at empowering Indian educators in robotics labs.
NEW · Updated Jul 13, 2026 · GoogleEmpowering India’s next generation of innovators with ATL Saathi
Google and AIM launched ATL Saathi, a Gemini-powered AI tool for Indian educators in robotics labs.
NEW · Updated Jul 13, 2026 · mAIEnergyA Multimodal Dataset for Large Language Model Applications in the Energy Domain
A paper introduces the mAIEnergy dataset, an open-access multimodal corpus for LLM applications in the energy domain, containing ~50k text documents, 20k images, 25M time series records, and 2M geospatial/relational entries, covering policy, scientific articles, satellite imagery, electricity measurements, weather data, and energy infrastructure. The dataset is FAIR-compliant and available on arXiv.
NEW · Updated Jul 11, 2026 · OpenAIGPT-5.6 launch: OpenAI pushes the model upgrade toward long-horizon autonomous agents
OpenAI releases GPT-5.6 and simultaneously launches ChatGPT Work for cross-app, file and long-term tasks.
NEW · Updated Jul 10, 2026 · Evolution of Accuracy and Visual-Cognitive Errors in a Decade of Vision-Language AI ModelsEvolution of Accuracy and Visual-Cognitive Errors in a Decade of Vision-Language AI Models
Vision language models (VLMs) have made remarkable progress in visual reasoning during the last decade. Most evaluations have used simple scenes (MS-COCO) that do not showcase complex human interactions or behaviors, only a handful of non-curated human descriptions as a benchmark, and have not focused on understanding visual-cognitive errors.
NEW · Updated Jul 10, 2026 · ConceptSMILEConceptSMILE: Auditing the Trustworthiness of Concept-Based Explainable AI
ConceptSMILE is a model-agnostic perturbation-based auditing framework for evaluating the reliability of concept-based explanations in AI.
NEW · Updated Jul 10, 2026 · ALICEALICE: Learning a General-Purpose Pathology Foundation Model from Vision, Vision-Language, and Slide-Level Experts
ALICE is a unified foundation model for computational pathology, trained through multi-stage agglomerative distillation from vision, vision-language, and slide-level experts. The paper is published on arXiv.
NEW · Updated Jul 10, 2026 · SAGEAgentSAGEAgent: A Self-Evolving Agent for Cost-Aware Modality Acquisition in Multimodal Survival Prediction
SAGEAgent is a self-evolving agent for cost-aware modality acquisition in multimodal survival prediction, as described in a paper on arXiv cs.AI published on 2026-07-10.
NEW · Updated Jul 10, 2026 · Deutsche Telekom, OpenAIHow Deutsche Telekom is rewiring telecommunications with AI
Deutsche Telekom is partnering with OpenAI to apply AI to customer service, employee workflows, network operations, and voice interactions, transforming into an AI-native telecom company.
NEW · Updated Jul 10, 2026 · Hugging FaceProfiling in PyTorch (Part 3): Attention is all you profile
Hugging Face released a blog post on 2026-07-10 titled 'Profiling in PyTorch (Part 3): Attention is all you profile', introducing profiling methods for attention mechanisms in PyTorch.
NEW · Updated Jul 10, 2026 · OpenAIGetting started with ChatGPT
OpenAI released a beginner's guide titled 'Getting started with ChatGPT' on its Academy platform on 2026-07-10, introducing how to start using ChatGPT for writing, brainstorming, and problem-solving.
NEW · Updated Jul 9, 2026 · OpenAIGPT-5.6 is now the preferred model in Microsoft 365 Copilot
OpenAI announced that GPT-5.6 has become the preferred model in Microsoft 365 Copilot, applied in Word, Excel, PowerPoint, Chat, and Cowork.
NEW · Updated Jul 9, 2026 · OpenAIChatGPT is now a partner for your most ambitious work
OpenAI released ChatGPT Work on 2026-07-09, an agent that can perform operations across applications and files, completing projects lasting several hours.
NEW · Updated Jul 9, 2026 · OpenAIGPT-5.6: Frontier intelligence that scales with your ambition
OpenAI released GPT-5.6 on 2026-07-09, claiming higher token intelligence, stronger performance per dollar, and on-demand scaling capabilities.
NEW · Updated Jul 9, 2026 · OpenAIGPT-5.5 Bio Bug Bounty
OpenAI launched a bug bounty program focused on biological risks associated with GPT-5.5, announced on 2026-07-09.
NEW · Updated Jul 8, 2026 · OpenAIOur approach to government and national security partnerships
OpenAI published an article on 2026-07-08 outlining its principles for partnering with government and national security agencies, including responsible use, democratic accountability, and public safety.
NEW · Updated Jul 8, 2026 · OpenAISeparating signal from noise in coding evaluations
OpenAI released an analysis pointing out reliability issues with the SWE-Bench Pro coding benchmark.
NEW · Updated Jul 8, 2026 · OpenAIHelping K–12 educators build practical AI skills
OpenAI Academy, in partnership with the Walton Family Foundation, is hosting hands-on AI skills workshops for K-12 educators to help them master practical AI skills for the classroom.
NEW · Updated Jul 8, 2026 · Hugging FaceNative-speed vLLM transformers modeling backend
Hugging Face publishes blog post announcing native-speed transformers modeling backend for vLLM.
NEW · Updated Jul 8, 2026 · OpenAIIntroducing GPT-Live
OpenAI released GPT-Live on 2026-07-08, a new generation voice model for natural voice interaction, integrated into ChatGPT Voice.
NEW · Updated Jul 7, 2026 · Hugging Face, Amazon SageMaker StudioFrom Hugging Face to Amazon SageMaker Studio in one click
Hugging Face integrates with Amazon SageMaker Studio, enabling one-click model deployment from Hugging Face model pages to SageMaker Studio.
NEW · Updated Jul 7, 2026 · Hugging Face, MicrosoftHugging Face Models on Foundry Managed Compute
On July 7, 2026, Hugging Face announced in a blog post that its models are now available on Microsoft Foundry Managed Compute.
NEW · Updated Jul 7, 2026 · Hugging Face, SkyPilotRun AI workloads on any cloud, store on Hugging Face: zero-egress storage with SkyPilot
Hugging Face partners with SkyPilot to offer zero-egress storage, allowing users to run AI workloads on any cloud and store data on Hugging Face without data egress fees.
NEW · Updated Jul 7, 2026 · MUFGMUFG aims to become AI-native with OpenAI
MUFG uses ChatGPT Enterprise to build an AI-native organization, improve workflows, and deliver new AI-powered financial services at scale.
NEW · Updated Jul 7, 2026 · Hugging FaceLeRobot v0.6.0: Imagine, Evaluate, Improve
Hugging Face released LeRobot v0.6.0 on 2026-07-07, with the version theme 'Imagine, Evaluate, Improve'.
NEW · Updated Jul 7, 2026 · Australian Payments PlusAustralian Payments Plus accelerates with ChatGPT and Codex
Australian Payments Plus (AP+) uses ChatGPT Enterprise and Codex to speed up payment system development, saving time, improving quality, and keeping human judgment central.
Updated Jul 6, 2026 · PhotoroomPRX Part 4: Our Data Strategy
Photoroom publishes Part 4 of the PRX series on Hugging Face blog, introducing its data strategy.
Updated Jul 3, 2026 · Google DeepMind, A24Google DeepMind and A24 Announce First-of-Its-Kind Research Partnership
Google DeepMind and independent film company A24 announced an unprecedented research partnership on July 3, 2026.
Updated Jul 1, 2026 · Hugging Face, Cerebras, Gemma 4Hugging Face and Cerebras bring Gemma 4 to real-time voice AI
Hugging Face and Cerebras collaborate to apply the Gemma 4 model to real-time voice AI.
ScarfBench: Benchmarking AI Agents for Enterprise Java Framework Migration
IBM Research released ScarfBench on Hugging Face, a benchmark for evaluating AI agents on enterprise Java framework migration tasks.
Updated Jun 30, 2026 · Google DeepMindStart building with Nano Banana 2 Lite and Gemini Omni Flash
Google DeepMind published a blog post on 2026-06-30 announcing that developers can start building with Nano Banana 2 Lite and Gemini Omni Flash.
Updated Jun 30, 2026 · Hugging FaceWhy Specialization Is Inevitable
Hugging Face published a blog post titled 'Why Specialization Is Inevitable' on 2026-06-30.
Updated Jun 30, 2026 · OpenAIHow ChatGPT adoption has expanded
OpenAI released Signals data on June 30, 2026, showing global ChatGPT adoption growth, increased user usage, exploration of more features, and growth across regions and languages.
Updated Jun 30, 2026 · OpenAICore dump epidemiology: fixing an 18-year-old bug
OpenAI engineers used large-scale core dump analysis to debug rare infrastructure crashes, uncovering both a hardware fault and an 18-year-old software bug.
Updated Jun 30, 2026 · OpenAIIntroducing GeneBench-Pro
OpenAI released GeneBench-Pro on 2026-06-30, a benchmark for evaluating AI performance in genomics, biology, and scientific research using complex real-world datasets.
Updated Jun 30, 2026 · Hugging FaceFeaturing Every Eval Ever Results on Hugging Face Model Pages
Hugging Face published a blog post on June 30, 2026, announcing the display of Every Eval Ever results on model pages.
Updated Jun 29, 2026 · Allen AIDiScoFormer: One transformer for density and score, across distributions
Allen AI published a blog post on Hugging Face introducing DiScoFormer, a single Transformer model that estimates both density and score across distributions.
Updated Jun 29, 2026 · OpenAIMapping Europe’s AI Workforce Opportunity
OpenAI released a report on June 29, 2026, mapping how AI reshapes EU employment, identifying occupations at risk of automation, growth, or workflow changes.