Z.ai’s GLM-5.3 goes open weight

Earlier in August, Z.ai, the Chinese AI lab behind the viral ox-alpha model that turned out to be GLM-5.3-Flash, launched its flagship GLM-5.3 model. On Friday, the company made the model’s weights available on Hugging Face, which Nvidia may soon own. Several third-party inference services already host it and make it available on services like OpenRouter, which Stripe will soon own.

As The New Stack’s Amanda Caswell reported when the model originally launched, Z.ai’s focus on training GLM-5.3 was on post-training. The result isn’t simply a large jump in benchmark performance over its predecessor, GLM-5.2, but a model that is often ahead of other Chinese open-weight models and can keep pace with current models from the large U.S. frontier labs.

This Is How Forward Deployed Engineering Is Actually Done

AI LABS explores the rising demand for forward-deployed engineers who bridge the gap between complex business processes and practical AI implementation. The session outlines a five-step framework for auditing existing workflows, identifying automation opportunities, and building systems that integrate seamlessly into a company's current operations without disrupting established productivity.

Qwen3.8-27B - 200 Tokens per Second

In this video, I look at the long awaited Qwen3.8-27B model.  Both what it can do and how to serve it at the maximum tokens per second

New v1.2 Skills!

Skills v1.2.0 is out with major improvements: new documentation site, Claude Code marketplace integration, five brand new skills including Wait What for Opus verbosity, an updated Grill Me with multi-question rounds, and the powerful Wizard skill for infrastructure provisioning.

Qwen3.8-27B runs frontier-class coding agents & reasoning locally

The biggest AI model release of the past few days, at least among the developers and AI power users on social media, wasn't a frontier cloud model from OpenAI, Anthropic or Google.

It was a 27-billion-parameter model from Alibaba: Qwen3.8-27B landed on Hugging Face on Friday under an enterprise-friendly, open source Apache 2.0 license, giving developers downloadable weights for a dense multimodal model.

This Claude Skill Just Fixed Loop Engineering

AI LABS explores the mechanism behind the popular Gauntlet Loop, a technique for building complex applications in one shot. By integrating Wayfinder, an intensive planning skill, AI LABS demonstrates how to overcome the method's inherent limitations regarding quality verification and project drift, providing a more structured approach to AI-driven software development.

Local AI On Apple Silicon uses 7X Less RAM

Better Stack explores how Turbo Fieldfare utilizes the unique architecture of Apple Silicon to run massive mixture-of-experts models. By streaming specific model components directly from SSD and leveraging unified memory, this approach significantly reduces RAM requirements, making advanced AI performance more accessible on local hardware.

OpenAI Just Announced The New Plugin Standard for AI Agents

The skills you have already written don't have to stay locked to one coding agent. A new open standard called Agent Plugins lets you bundle Agent Skills and MCP servers into a single folder that works across Codex, ChatGPT, Cursor, GitHub Copilot, Kiro and VS Code, with Google joining more recently.

A plugin under this standard is just a directory. Your skills folder, exactly as you already have it, a plugin.json manifest, and an mcp.json where your MCP servers live. That is the portable core, and it is a much simpler file structure than you would expect from a spec with this many big names on it.

AI code-testing startup Blacksmith’s valuation jumps almost 10x

As AI makes coding dramatically faster, the next big challenge in software development is testing and validating all that code. Blacksmith has raised a new $45 million round to capitalize on that shift.

Founded in 2024, Blacksmith helps companies build, test, and verify software before it reaches production. The startup now serves more than 5,000 customers, including Mercury, Supabase, Clerk, Ashby, and Expensify, up from more than 700 customers less than a year ago, co-founder and CEO Aditya Jayaprakash said in an exclusive interview.

AWS Continuum integrates with Codex and Claude Code

Amazon Web Services is threading its AI-powered security infrastructure directly into the coding environments built by two of its fiercest rivals — and in doing so, it is making a bold bet that controlling the security layer matters more than controlling the model.

AWS announced at Black Hat USA 2026 this month that its Continuum platform for code vulnerabilities will integrate directly into Anthropic's Claude Code and OpenAI's Codex, alongside AWS's own Kiro IDE.

Kestra IS THE N8N KILLER!

WorldofAI demonstrates how to integrate Kestra with AI agents to create reliable, human-in-the-loop automation workflows. By leveraging YAML-based configurations and native integrations with services like GitHub, users can manage predictable tasks alongside AI-driven reasoning, ensuring critical actions only proceed after manual review and approval.

LFM2.5-2.6B model brings powerful AI agents to small devices

Earlier this week, the AI startup Liquid, formed in 2023 by former MIT computer scientists, debuted LFM2.5-2.6B, a new open-weight language model designed specifically for agentic workloads. 

In release materials and a recent interview with VentureBeat, Liquid's researchers said LFM2.5-2.6B can run entirely on local hardware — from smartphones and laptops down to a Raspberry Pi — without relying on cloud inference or GPUs, unlocking edge AI applications and giving more options to enterprises working in regulated industries or with sensitive information they don't want to send up to the cloud.

AI startup Hark unveils first product

Hark, the secretive AI startup founded earlier this year by serial entrepreneur and roboticist Brett Adcock, today announced Handoff, a "computer use agent" (CUA) that it says is among the top-performing in the world at navigating the open web on a user's behalf — ordering dinner on DoorDash, booking flights on United and Delta, or messaging job candidates on LinkedIn — all autonomously, end-to-end.

Qwen 3.8 Max IS OUT!

WorldofAI explores the capabilities of this large-scale model, focusing on its performance in coding, research, and long-horizon reasoning. The review examines its multimodal functionality and utility in generating 3D assets, web applications, and interactive simulations, while demonstrating the test-sprite CLI tool as a mechanism for verifying agent-driven code outputs.

Your agent needs a computer, not a container

The @cloudflare/computer package provides an agent runtime where the details and mechanics of what code runs in an isolate, a container sandbox, or a web browser are handled by the platform. Each agent gets a computer, the runtime optimizes for efficiency, and scalability.

Claude Code Creator's Greatest Tip For Using AI Agents

Boris Churnney explains strategies for optimizing AI agent workflows, emphasizing why clearing outdated setup files and prompts is crucial when new models are released. Learn how to refine project management by implementing strict evaluation checks, utilizing effective coding habits, and balancing automation with manual oversight to ensure better results from intelligent systems.