Blog

NVIDIA introduces the Agent Toolkit, an open modular foundation with models, tools, skills, and a secure runtime to help businesses build specialized, trustworthy AI agents for various industries.

IBM introduces CUGA, an open-source agent harness that handles plumbing for state, tool calls, and orchestration, allowing developers to focus on defining tools and prompts. The article showcases two dozen single-file example apps built with CUGA, demonstrating how it eliminates repetitive framework setup.

NVIDIA technology now powers over 400 of the world's 500 fastest supercomputers (81% of the TOP500), with record GPU and networking adoption and top efficiency on the Green500 list.

NVIDIA announces new AI agents and tools for telecom operations, including synthetic data generation and secure agent runtimes, showcased at DTW Ignite 2026. The platform aims to enable autonomous networks by combining domain-specific models, privacy-safe synthetic data, and policy-based guardrails.

MIT researchers have developed a new system-on-a-chip that enables tiny robots to create detailed 3D maps of their environments in real-time using only about 6 milliwatts of power, potentially enabling long-duration autonomous navigation in complex spaces.

Tencent is testing an AI assistant called Xiaowei within its WeChat app in China, aiming to catch up with rivals in the AI market by leveraging its massive user base.

Anthropic is extending its Cowork agentic system to mobile apps, with evidence of cloud-based execution to remove the need for a desktop machine to remain awake, along with groundwork for a voice model refresh.

Anthropic partner provider shows slug for upcoming Claude Sonnet 5 model, hinting at imminent release.

Anthropic updated its privacy policy to require some flagged Claude users to upload government-issued ID for identity verification, as part of an appeals process to avoid account bans, amid regulatory and White House pressures.

An analysis of AI model size scaling trends from 2023 to 2031, published on LessWrong.

GLM-5.2 is a new open-source AI model that sets a high bar for open models, though it still trails proprietary frontier models and lacks some features like vision.

The article presents 'knowledge agents', a methodology that injects relevant knowledge into AI agents via a hybrid retrieval system, allowing smaller models to outperform large frontier models across specialized domains like financial markets, policy, and healthcare.

Alibaba released HappyHorse 1.1, a major AI video generation model upgrade now available via API, rising to No. 2 in global rankings as competitors Sora and Seedance faltered.

OpenAI launches new security tools including Codex Security plugin and an updated GPT-5.5-Cyber model, alongside the Daybreak initiative and Patch the Planet open-source project, shifting from vulnerability discovery to automated patch generation.

SpaceX has signed a computing power deal with open-source AI startup Reflection, worth up to $6.3 billion, giving Reflection access to Nvidia GB300s via SpaceX's Colossus data center. The deal highlights SpaceX's expansion into selling computing capacity and the growing momentum of open-source AI.

Hugging Face describes how they built a weekly release pipeline for their huggingface_hub library using AI, open-source tools, and human oversight, enabling faster and more reliable releases.

Omio is using OpenAI's ChatGPT and Codex to build conversational travel booking experiences and transform internal operations, moving toward an AI-native approach.

Research paper shows that LLMs suffer from 'role confusion', where they prioritize the style of text over its actual role tags, enabling prompt injection attacks. Destyling text reduces attack success from 61% to 10%, indicating a fundamental challenge for LLM security.

Simon Willison ported the Moebius 0.2B image inpainting model to run in the browser using WebGPU and ONNX Runtime, assisted by Claude Code. The resulting demo allows users to upload images and remove objects via inpainting.

PP-OCRv6 is the latest generation of PaddleOCR's universal OCR model family, offering three tiers from 1.5M to 34.5M parameters, supporting 50 languages, and achieving significant accuracy improvements over previous versions.