xAI Launches Grok Bot and NVIDIA Drops Open Nemotron 3.5 Lightning Model

Today's AI news: xAI launched Grok Bot, always-on AI teammates that log into your existing tools and finish real work unsupervised, NVIDIA released an open 30B Nemotron 3.5 Lightning model built for fast agent execution, OpenAI synced ChatGPT Work and Codex so you can import projects, chats, skills and plugins between agents, and LTX shipped a major upgrade to its LTX-2.5 world model. Chapters: 0:00 Today's AI News 0:29 Grok Bot 1:20 NVIDIA Nemotron 3.5 Lightning 2:07 OpenAI ChatGPT Work Codex Sync 2:47 LTX-2.5 World Model Blend Roundup 2026-08-11 https://x.com/bot/status/2087224798078517251 https://x.com/NVIDIAAI/status/2087162151995629926 https://x.com/OpenAIDevs/status/2087242829076791392 https://x.com/ltx_io/status/2087255203489755243 [quick] xAI launched Grok Bot, AI teammates that log into your tools and finish real work, NVIDIA released an open 30B Nemotron 3.5 Lightning for fast agent tasks, OpenAI synced ChatGPT Work and Codex across agents, and LTX shipped LTX-2.5, a major world-model upgrade. [excited] Here's today's AI news. [quick] xAI has launched Grok Bot, a new always-on AI teammate that signs into your existing tools and works through a task much like a human operator would, navigating websites, entering information, and carrying the job through to completion without waiting for another prompt. Multiple Bots can work together on the same project, sharing context in a group chat with one Bot coordinating the others, and you can demonstrate a workflow once for a Bot to save as a reusable routine. It's in beta for SuperGrok Heavy and Cursor Ultra and Teams Premium subscribers on desktop and iOS, and early users are already having it negotiate with vendors and keep CRMs updated on its own. It's supposed to be less of a chatbot you supervise and more of a colleague you delegate to. Let's see how it goes. [quick] NVIDIA has released Nemotron 3.5 Lightning, an open 30-billion-parameter model with just 3 billion active parameters, built specifically for the execution layer of always-on AI agents handling high-volume, specialised tasks. It uses a hybrid architecture blending Mamba and mixture-of-experts layers, supports a million-token context window, and delivers up to four times the output speed of similarly sized models while completing tasks 30 percent faster. It's released under a permissive open licence and ready for commercial use, another sign that the fight for the agent execution layer, not just the flagship reasoning models, is heating up fast. [quick] OpenAI is now letting you keep your work from other AI agents in sync with ChatGPT Work and Codex, importing projects, chats, skills and plugins from elsewhere and opting into automatic updates from Settings. It's available now in the ChatGPT desktop app, which brings Chat, Work and Codex together in one place, and it means switching agent tools no longer means starting your setup from scratch. For anyone juggling multiple AI coding and research tools day to day, that's a real reduction in the busywork of keeping them all in sync. [quick] LTX has shipped LTX-2.5, a major upgrade to the open-weights world model already powering film, robotics and real-time workflows, adding a better video decoder, native multi-shot scenes that stay consistent across cuts, and stronger prompt understanding. It now supports native 4K HDR and RAW output, automatic clip duration, and a faster distilled model that can generate a 10-second clip in under 7 seconds on two GB200 GPUs. It's available as open weights, through a managed API, and inside LTX Studio, so teams can build on the same foundation model however they prefer to deploy it.