Kimi K3 Launches on Databricks

Moonshot AI's flagship model arrives on Databricks, bringing enterprise-grade reasoning to the platform.

Moonshot AI's Kimi K3 is now available on Databricks for enterprise use. The integration brings one of China's leading large language models to the Databricks ecosystem, enabling enterprise customers to run advanced reasoning workloads directly within their data infrastructure. Kimi K3 has been recognized for its strong performance on coding, long-context reasoning, and multilingual tasks.

Hy3D WorldClaw: Generate 3D Worlds from Text Prompts

Tencent Hunyuan unveils an agentic workflow that builds explorable, game-ready 3D open worlds from text.

Tencent Hunyuan unveiled Hy3D WorldClaw, which generates large-scale, freely explorable 3D open worlds from text prompts, built entirely from editable game-grade assets. Unlike video or Gaussian Splatting approaches, WorldClaw produces structured 3D environments that users can freely navigate and modify, marking a significant step toward production-ready AI-driven worldbuilding for games and simulations.

ChatGPT Desktop App Supports Cross-Agent Import Sync

Import projects, chats, skills, and plugins from other agents with automatic sync updates.

OpenAI now allows users to import projects, chats, skills, and plugins from other agents, with automatic sync updates available in the desktop app. Users can review import history and opt in to automatic updates in Settings, enabling seamless workflows across multiple AI coding assistants.

Codex Lands on Linux with ChatGPT Desktop

Build where your code already lives: Codex is coming to the ChatGPT Linux desktop app.

OpenAI announced that Codex will arrive on the Linux version of the ChatGPT desktop app, letting developers build directly where their code lives. The integration keeps projects, development workflows, and supported browser tools together in one native application.

SGLang Ships Day-0 Support for Nemotron 3.5 Lightning

SGLang launched Day 0 support for NVIDIA Nemotron 3.5 Lightning, a 30B hybrid Mamba-Transformer MoE with just 3B active parameters, distilled from Nemotron 3 Ultra for always-on agents. The release includes three built-in speculators — MTP, DFlash, DSpark — and native support for coding agents, tool use, and agent harnesses.

Nemotron 3.5 Lightning Now Available on Ollama

NVIDIA's 30B model for always-on agents is now on Ollama, running entirely local. Launch commands for Claude Code, Hermes Agent, and OpenClaw are included out of the box.

Nemotron 3.5 Lightning on LM Studio

The 30B MoE model with 3B active parameters runs fast on LM Studio, supporting up to 1M context window. Distilled from Nemotron 3 Ultra for high-volume agentic use cases with tool use support.

Gemini App Surpasses 1 Billion Monthly Users

Google's fastest-growing product ever and the company's 14th to hit the billion-user mark.

CEO Sundar Pichai announced that over one billion people now use the Gemini app every month, making it Google's fastest-growing product ever and its 14th to cross the billion-user threshold. Pichai credited Josh Woodward and the entire Gemini team for the achievement.

US Open AI Model Development Is Back

Graham Neubig of CMU notes that back-to-back 30B releases from Meta and NVIDIA signal a resurgence in open US AI development. "This is great for local AI, given that most other labs have been building bigger and bigger models that are not easy to run locally."

Meta Muse Glimmer: First New Open-Weight LLM Since Llama Era

Meta released Muse Glimmer, a 30B multimodal reasoning model with a Gemma-like architecture. Weights are available on Hugging Face with BF16 weights, GGUF k-quants, ExecuTorch builds, and DFlash draft models.

Nathan Lambert: "We Don't Need Policy on Distillation, We Need Products to Work"

"When I was in China it was insinuated that every lab did this. I'm glad there's public research on it and am still shocked the frontier labs haven't patched this stuff. We don't need policy action on distillation, we just need the products to work as intended." The thread resonated widely with 677 likes and 66K views.