AI news — 16 November 2026

Every morning we collect the most important AI news from dozens of sources and sum it up in a few lines, with the newest open-source projects beside it. Each card links to the original source.

RSS · Today's newsRSS · New repositories

News 7 stories

The Decoder
Models & releases

Meituan releases LongCat-2.5 agent model trained on Chinese hardware

Chinese company Meituan has released the weights for LongCat-2.5, a 1.6-trillion-parameter model designed to handle complex coding and agent tasks autonomously. The model was trained entirely on domestic hardware including Huawei Ascend chips, achieving benchmark scores close to Claude Opus 5 without Nvidia processors. It supports multimodal inputs across text, images, and video, handling a context window of up to one million tokens.

Original

Meituan releases a 1.6-trillion-parameter agent model that never touched an Nvidia GPU

Meituan has released the weights for LongCat-2.5, a 1.6-trillion-parameter model built for AI agents and trained entirely on Chinese hardware. On Terminal Bench, it scores 88.4 points, landing just behind Claude Opus 5. The sample deployment runs on 194 Huawei Ascend chips, and Meituan makes no mention of Nvidia GPUs. The article Meituan releases a 1.6-trillion-parameter agent model that never…

OpenAI
Research

Defining the Standard Model on a Lattice with GPT-6 Astra

Researchers explored a candidate nonperturbative lattice definition of the Standard Model. The model GPT-6 Astra helped turn the initial concept into a detailed proof.

Original

A step toward defining the Standard Model on a lattice

Explore a candidate nonperturbative lattice definition of the Standard Model and how GPT-6 Astra helped turn the idea into a detailed proof.

Hacker News
Policy & safety

Developer Argues All AI-Generated Code Should Be Open Source

Developer Jakub T. Jankiewicz argues that all code generated with the help of artificial intelligence should be Open Source or released under a CC0 license. He notes that AI models were trained on scraped Open Source repositories and Stack Overflow data without required attribution or permissions.

Original

All AI-generated code should be Open Source

Hacker News
Research

Sebastian Raschka Explains Positional Embeddings in Modern LLMs

Sebastian Raschka published a technical guide explaining how large language models encode and process token position information. The walkthrough covers both traditional methods and relative positional embeddings like RoPE, accompanied by visual examples and a PyTorch implementation. It also serves as foundational material examining positional embedding strategies across open-weight models.

Original

Understanding Positional Embeddings in Modern LLMs

arXiv cs.AI
Research

Study Analyzes Coordination and Inference Compute in Multi-Agent LLMs

Researcher Seyed Mahmoud Sajjadi Mohammadabadi examined how multi-agent language models are evaluated, noting that comparing them to a single greedy pass confounds coordination with sample multiplicity and inference compute. This analysis helps untangle the actual benefits of agent coordination from increased test-time computation.

Original

More Agents, More Samples, or More Thinking? Untangling Coordination, Ensembling, and Test-Time Compute in Multi-Agent LLMs

arXiv:2611.17610v1 Announce Type: new Abstract: Evaluations of multi-agent language models often compare communicating architectures against a single greedy pass. This practice confounds coordination with sample multiplicity and inference compute. We synthesize nine primary empirical evaluations comprising 516 unique multi-agent configurations organized along a resource vector R with six…

arXiv cs.AI
Research

Auditing Hallucination Detectors on Small Quantized On-Device LLMs

Researchers Shubhra Mittal and Gaurav Naresh Mittal investigated hallucination detection methods on small, 4-bit quantized language models.

Original

Does the Ranking Hold? A Practitioner-Facing Audit of Hallucination Detectors on Small, Quantized, On-Device LLMs

arXiv:2611.17339v1 Announce Type: new Abstract: Hallucination detection for large language models is dominated by methods evaluated on 7B-70B research models with abundant sampling budgets and exact-match or ROUGE correctness labels. The small models practitioners actually ship on-device, Phi-3-mini, Llama-3.2-3B-Instruct, Qwen2.5-3B-Instruct, and Gemma-2-2B-Instruct, typically at 4-bit…

Open-source repositories 8 repositories

Trending on GitHub★ 19,357
Agents

andrewyng/openworker

OpenWorker provides a desktop application and an SDK for running and building agentic workflows for everyday work. It supports cloud or local models via Ollama, allowing users to automate research, writing, and code reviews with built-in permission controls and sandboxing.

  • python
  • MIT
  • #agents
  • #sdk
  • #desktop
Trending on GitHub★ 7,385
Coding

GoogleChrome/modern-web-guidance-src

This project equips coding agents with modern web platform best practices and browser compatibility data to steer them away from outdated code patterns. It injects token-efficient, expert-curated guidance directly into an agent's context window and includes tools to scan codebases for legacy implementations.

  • TypeScript
  • Apache-2.0
  • #coding
  • #agents
  • #web
New on GitHub★ 3,378
Agents

benjitaylor/skills

A collection of skills for AI coding agents focused on design and motion work. It provides tools and guidance, such as Stageshow, to help agents build realistic React interfaces and record the results to review their work.

  • TypeScript
  • #agents
  • #design
  • #react
Open repository about +161 stars a day
Trending on GitHub★ 1,881
Agents

NVIDIA/SkillEvaluator

NVIDIA SkillEvaluator is a multi-tier framework for evaluating AI agent skills and measuring how they impact agent behavior and task results. It allows developers to run quality checks directly via CLI or CI pipelines, supporting deterministic tests without an LLM as well as model-backed evaluations.

  • python
  • Apache-2.0
  • #agents
  • #evaluation
  • #testing
New on GitHub★ 1,649
Design & images

hitkeeen/blender-ai-mcp-addon

This project integrates an AI assistant and an MCP server directly into Blender with a visual chat workspace.

  • MIT
  • #blender
  • #mcp
  • #3d
Open repository about +456 stars a day
Hugging Face model♥ 419
Design & images

Comfy-Org/MiniMax-H3-Image

This project provides repackaged MiniMax H3 Image model files optimized for use within ComfyUI. It enables both text-to-image generation and image editing workflows alongside compatible text encoders and VAE components.

  • other
  • #text-to-image
  • #comfyui
  • #diffusion
Hugging Face model♥ 325
Audio & video

higurashi-ai/Haru-4B

Haru-4B is a multilingual full-duplex speech dialogue model designed to listen and speak simultaneously with very low latency. It generates continuous audio every 133 milliseconds and allows custom voice and character control using a text prompt and an audio reference.

  • apache-2.0
  • #audio
  • #speech
  • #voice
Paper with code★ 4
Audio & video

ZhouTimeMachine/JVT

JVT is a DiT-style Transformer that performs video diffusion directly in pixel space without requiring pretrained autoencoders or U-Net skip connections. It supports unconditional, class-conditioned, and camera-conditioned generation, including joint RGB-disparity outputs.

  • Python
  • #video-generation
  • #diffusion
  • #transformer

Summaries are written automatically from the sources' headlines and excerpts and may contain mistakes. The full story is always at the source. To have a link or source removed, write to us from the contact section.

Let's start

Have an idea or a system that needs transforming?

Write us a few lines. Your message reaches us instantly, and we usually reply within one business day.

scouthub@lab: ~/contact

scouthub@lab:~$ contact --new --lang=en