Latest Artificial Intelligence News & Dispatches

AG Kit (7,768 GitHub stars, 2026) is an open-source multi-agent orchestration framework with 20 pre-built agent personas (Frontend, Backend, Security, PM, QA, Data Engineer) and 13 interactive slash-command workflows. Ge...

Compendium (July 2026) is the first agent-native workspace where AI agents operate as equal team members with persistent context, profiles, and permission-based autonomy. Notion AI provides AI-powered writing assistance...

NVIDIA Puzzle-75B-A9B (arXiv 2607.04371, July 2026) is a compressed Mixture-of-Experts model achieving 2.03x throughput over the dense baseline through Neural Architecture Search (NAS), NVFP4 quantization, and expert mer...

Frugon (4,200+ GitHub stars, July 2026) is an open-source intelligent model routing engine that sends simple queries to cheap models and complex ones to frontier models. Portkey is a managed AI gateway with fallback stra...

AReaL (5,500+ GitHub stars, July 2026), TRL (Hugging Face, 8,000+ stars), and RL4LMs (research framework) are the three leading open-source frameworks for reinforcement learning with LLM agents. AReaL provides PPO, REINF...

Vercel Agent (July 2026) is Vercel's production agent monitoring platform with built-in canary deployments, auto-rollback, error rate detection, latency monitoring, and drift detection for AI agents. Datadog AI provides...

NemoClaw (LangChain+NVIDIA, July 2026) is a sandboxed enterprise agent blueprint combining Deep Agents Code, Nemotron 3 Ultra, and OpenShell. OpenClaw is an MCP-native agent runtime. Hermes is a cross-provider agent orch...

Microsoft Agent Skills for .NET (GA July 2026) and LangChain Agent Skills are two approaches to packaging domain expertise for AI agents. Microsoft Skills use a four-stage progressive disclosure pattern (advertise, load,...

GPT-Live-1 (OpenAI, July 2026), Gemini Live (Google, March 2026), and Seeduplex (ByteDance, April 2026) are the three leading full-duplex voice AI models. GPT-Live uses a two-layer architecture with GPT-5.5 backend deleg...

Grok 4.5 (SpaceXAI/Cursor, July 2026) is a mixture-of-experts model trained on trillions of Cursor interaction tokens. It costs $2/M input tokens vs Opus 4.8 at $5/M and GPT-5.5 at $5/M. On SWE-Bench Pro, Grok 4.5 uses 4...

Refiant Protea (July 2026) is a suite of long-context AI models offering 1M, 5M, and 10M token context windows using swarm optimization algorithms. It solves the lost-in-the-middle problem and processes entire enterprise...

AMD GAIA (v0.21.2, MIT License, 2026) is AMD's open-source framework for building AI agents that run 100% locally on AMD Ryzen AI hardware. It provides agent orchestration, tool registry, RAG over 50+ file formats, deskt...

NVIDIA Audex 30B-A3B (July 2026) and Qwen3.5-35B-A3B are the two leading open audio-text LLMs. Audex uniquely handles both audio understanding and generation in a single model while preserving text intelligence. Qwen3.5-...

Microsoft Foundry Autopilot Agents (public preview June 2026) are AI agents with their own Entra Agent ID, productivity license, email, calendar, OneDrive, and Teams presence. They work in shared spaces like Teams group...

OpenAI GPT-Realtime-2.1 is a speech-to-speech reasoning model for building low-latency voice agents with configurable reasoning effort, 128K context, tool calling, and multi-agent handoffs. Unlike chained ASR-LLM-TTS arc...