News Archive

NVIDIA research addresses robotic generalization and autonomous driving safety through improved gripper control and agent training methods at scale.

NVIDIA unveils physical AI agent skills for autonomous vehicles, robotics, and vision AI, enabling developers to accelerate research by providing full workflow tools beyond model development.

Marketing content about Google Search features for shopping; not developer-relevant.

Research on Direct Preference Optimization techniques applicable to model training beyond conversational AI.

Service incident affecting Codex, ChatGPT, and Responses API has been resolved with all components operational.

Service incident on Claude Opus 4.7 with elevated error rates has been resolved after implementing a fix.

Claude Code services experienced degradation affecting security reviews, code reviews, routines, and web sessions. Engineers were actively investigating and resolving the issue.

Service incident affecting ChatGPT Pro and related services has been resolved with full recovery.

Service incident with Codex API returning invalid model reference errors has been resolved.

Hugging Face adds MCP (Model Context Protocol) tool support to Reachy Mini robotic platform.

NVIDIA NemoClaw enables industrial software vendors to build autonomous AI engineers that optimize engineering workflows from CAD through post-processing.

NVIDIA and Microsoft partner on unified agentic AI stack spanning Windows, cloud, and local deployment with optimized hardware, runtimes, and long-context reasoning models.

Microsoft and NVIDIA provide new tools for developers to build personal AI agents running on Windows PCs.

NVIDIA enables self-evolving agents using Hermes Agent and NemoClaw for research data synthesis and acceleration with improved security.

Holo3.1 enables fast, local execution of computer use agents without external dependencies, allowing developers to build autonomous task-completion systems that run on-device.

Service incident with elevated error rates for guest users accessing ChatGPT conversations has been resolved.

Anthropic experienced elevated errors across multiple models on June 2nd, identified and being fixed. Service was briefly unavailable during the incident.

NVIDIA highlights how financial institutions are adopting transaction foundation models to unify fragmented AI systems across fraud detection, credit, recommendations, and risk.

NVIDIA announces JetPack 7.2 and NemoClaw support on Jetson platform, enabling agentic AI deployment on edge devices with improved performance on Jetson Orin.

NVIDIA JetPack 7.2 brings memory-efficient agentic AI deployment to edge devices for physical world applications.

NVIDIA DGX Spark enables local AI agent deployment with optimized models, large context windows, and multi-node clustering for concurrent long-running tasks.

Claude Sonnet 4.6 experienced elevated error rates which have been resolved. Service is now operating normally.

Service incident with decreased ChatGPT availability for free-tier users has been resolved.

Google describes how Gemini was used in the development and presentation of Google I/O 2026 event content.

JetBrains released Mellum2, a 12B parameter mixture-of-experts model available on Hugging Face.

Claude Sonnet 4.6 experienced elevated error rates that were identified, fixed, and resolved within ~45 minutes on Jun 1.

Article discussing enterprise AI adoption strategies beyond large language models, focusing on agent-based approaches.

Claude Opus 4.7 experienced elevated error rates that were identified and resolved with a fix on Jun 1.

Sonnet 4.5 experienced elevated errors that were identified, fixed, and resolved within ~20 minutes on Jun 1.

Opus 4.7 experienced elevated error rates that were identified, fixed, and resolved within ~35 minutes on Jun 1.

NVIDIA AI Cloud ecosystem expands globally with partner infrastructure to support enterprise and developer demand for agentic AI compute resources.

NVIDIA announces Factory Operations Blueprint, an AI system connecting machine signals and operational data into unified decision layer for plant-wide manufacturing intelligence.

Taiwan's manufacturers expand NVIDIA ecosystem infrastructure production, including MGX rack components for Vera Rubin platforms powering agentic AI factories.

NVIDIA Alpamayo enables closed-loop post-training of autonomous vehicle vision-language-action models, bridging training and deployment.

NVIDIA Cosmos 3 helps physical AI systems reason about the world before taking action.

NVIDIA released Cosmos 3, an open omni-model for physical AI reasoning and action tasks.

NVIDIA Cosmos 3 provides tools for developing physical AI reasoning, world models, and action models for robots and autonomous systems.

NVIDIA DOCA in-silicon security advances AI factory infrastructure for agentic AI deployment with improved security for autonomous agents.

NVIDIA Vera CPU is optimized for agentic AI workloads in AI factories, setting new performance standards for long-context token generation.

NVIDIA DSX OS provides open modular software for operating AI factories at scale, enabling token generation and agent deployment.

OpenRouter released speech APIs, model fusion, and 20 new models including Gemini 3.5 Flash and Claude Opus 4.8 with enterprise workspace controls.

Opus 4.7 experienced elevated errors that were resolved on May 31 after investigation.

DynoSim addresses LLM serving optimization challenges by simulating performance trade-offs across model backends, tensor parallelism, and prefill/decode configurations.

Claude Opus 4.8 experienced elevated error rates for ~26 minutes on May 29; a fix was deployed and success rates returned to normal.

Google created an interactive quiz about I/O 2026 announcements using Google AI Studio.

Google published demonstrations of Gemini Omni and Gemini 3.5 models in action.

Service incident affecting ChatGPT access and multiple features has been resolved with all components operational.

OpenRouter released configurable security and governance tools including budget enforcement, data retention controls, model restrictions, and injection defense.

Google showcased AI prototypes from the Futures Lab developed by University of Waterloo students.

A trending quantized model variant (LFM2.5-8B-A1B-GGUF) by unsloth gained significant downloads and likes on the Hugging Face model hub.

Service incident with ChatGPT conversation issues has been resolved.

Service incident affecting OpenAI login and account creation has been resolved.

Service incident with business plan subscription checkout has been resolved.

Tutorial on PyTorch profiling using torch.profiler for performance analysis and optimization.

Google published a summary of 12 major announcements from Google I/O 2026.

OpenRouter raised $113M in Series B funding led by CapitalG with participation from multiple venture firms.

LiquidAI's LFM2.5-8B-A1B model trending on Hugging Face with high weekly downloads.

Service incident with elevated latency in Codex context compaction has been resolved.

NVIDIA's Qwen3.6-35B-A3B-NVFP4 model trending on Hugging Face with significant weekly download volume.

Reachy Mini robot platform now supports fully local AI inference capabilities.

Delta Weight Sync in TRL enables efficient distributed training of trillion-parameter models using Hub storage.

JetBrains' Mellum2-12B-A2.5B-Thinking model is trending on Hugging Face with 163 likes and 6,938 downloads this week.

Article discussing terminology and conceptual frameworks for AI agents and scaffolding approaches.

LiquidAI's LFM2.5-8B-A1B-GGUF model is trending with 165 likes and 87,045 downloads this week.

stepfun-ai's Step-3.7-Flash model is trending with 222 likes and 17,965 downloads this week.

Google I/O 2026 conference event featuring discussion with Alphabet CEO Sundar Pichai.

openbmb's MiniCPM5-1B model is trending with 745 likes and 68,494 downloads this week.

Google announces community investments and workforce development programs in Missouri.

Summary of 100 announcements from Google I/O 2026 conference.

Google Beam adds experimental group meeting features for hybrid video conferencing.

OpenRouter evaluates 11 LLMs including Claude and Grok across robotic tasks, demonstrating significant performance differences in model selection for physical AI.

sapientinc's HRM-Text-1B model is trending with 510 likes and 155,558 downloads this week.

NemoStation's Marlin-2B model is trending with 508 likes and 18,315 downloads this week.

OpenRouter Agent SDK adds human-in-the-loop tool type enabling auto-resolution of routine decisions with human input for high-stakes choices.

OpenRouter enables web search and page fetching capabilities for all tool-calling models across multiple search and fetch engine providers.

GPT-5.5 token prices doubled on OpenRouter; actual usage cost impact is reduced due to the model's decreased verbosity.

OpenRouter launches text-to-speech and audio transcription APIs supporting multiple providers under a unified interface.

Response Caching feature on OpenRouter caches identical requests for zero-cost retrieval with significantly faster response times.

OpenRouter April release includes video generation, workspaces, Agent SDK, reranker models, and frontier model launches.

DeepSeek's DeepSeek-V4-Pro model is trending with 4,590 likes and 5.8M downloads this week.

OpenRouter's Generation API endpoint was inaccessible on April 14 and has been restored.

Groq resolved performance issues affecting openai/gpt-oss-120b model; service is operating normally.

Vertex AI Gemini API experienced elevated error rates on the global endpoint from Feb 27, 04:37-06:35 UTC; issue resolved.