NVIDIA Nemotron 3 Ultra: 550B MoE Open-Weights Model for Agents
NVIDIA unveils Nemotron 3 Ultra, a 550B hybrid Mamba+Transformer MoE with open weights, faster inference, lower cost, and long-horizon agent planning.
Articles
X Launches Grok Build: Grok CLI Agent for Terminal Automation
X introduces Grok Build, a terminal-based CLI agent that edits files, runs commands, supports MCP servers, and uses parallel sub-agents to plan and execute tasks.
Anthropic Preps Public Launch of New Mythos as Oceanus Leaks
Anthropic’s Oceanus checkpoint hit red teamers ahead of a possible Mythos release, but the program was suspended after alleged resale via a Chinese API proxy.
Microsoft Scout: Autopilot AI Agent for Teams, Outlook & OneDrive
Microsoft introduced Scout, its first Autopilot AI agent that runs in the background to move tasks forward across Teams, Outlook, OneDrive and more.
Microsoft Launches OpenClaw Client with MXC Secure Containers
Microsoft released an official OpenClaw client with permissions, tray control, and secure containerized launch using Microsoft Execution Containers (MXC).
Microsoft Launches MAI AI Models for Copilot: Code, Reasoning, Image
Microsoft introduces in-house MAI models for Copilot, including MAI-Code-1-Flash, MAI-Thinking-1, MAI-Image-2.5 and MAI-Transcribe-1.5.
Claude Dynamic Workflows: New “ultracode” Trigger Replaces “workflow”
Claude now uses “ultracode” to reliably trigger dynamic workflows. “Workflow” may still work, but won’t auto-run if Claude thinks it’s not needed.
Hermes Desktop Launches: GUI, Remote Gateway, and Shared Context
Hermes Desktop is now in public preview for macOS, Windows, and Linux—bringing a full GUI, shared sessions with the CLI, and a VPS Remote Gateway.
Get 6 Months of Free ChatGPT Pro: Codex for Open Source Program
OpenAI’s Codex for Open Source offers maintainers 6 months of ChatGPT Pro, Codex access, API credits, and more. Apply in minutes with your GitHub repo.
Perplexity Launches Search as Code for Faster Agent Search
Perplexity’s Search as Code lets agents generate Python to run parallel queries, dedupe, filter, and rank results—cutting latency and costs on benchmarks.