01What Is GLM 5.2 and How It Works

GLM 5.2 is a Chinese AI coding model that can build and automate anything you want. You describe what you need in plain English, and it writes the code, creates the graphics, and even makes full 3D games. It runs on a coding plan and gives you plenty of tokens so you can work on large projects without running out.

In this episode, Julian Goldie shows how he used GLM 5.2 to build a full operating system, a 3D game, and many other tools in just one morning. The model is open source and MIT licensed, which means anyone can download it from Hugging Face and run it locally if they have a powerful setup. It competes with top models like Opus 4.8 and GPT 5.5, and even beats Google's Gemini 3.1 Pro on several benchmarks.

02GLM 5.2 Main Features and Benchmarks

GLM 5.2 has frontier-level intelligence. It outperforms GPT 5.5 on SWE, post-train bench, and humanities last exam. It matches Opus 4.8 on SWE and humanities.

Julian says these results changed his mind about open source models.

  • SWE benchmark: GLM 5.2 scores as high as Opus 4.8 and beats GPT 5.5.
  • Post-train bench: Outperforms GPT 5.5.
  • Humanities last exam: Right up there with Opus 4.8.
  • Outperforms Google's Gemini 3.1 Pro across almost all tests.

The model can build websites, games, apps, and tools. It also works with tool use, so it can interact with outside systems. Because it's open source, anyone can improve it, and it can self-iterate to get better over time.

03How to Set Up GLM 5.2 in Three Ways

  1. Use the CLI dashboard: Go to GLM 5.2 Insider 8. Type what you want in plain English. The model codes it and places the result in your workspace. No technical skill needed.
  2. Plug into Hermes agent: Run the one-click install command for Hermes. Then type hermes model in your terminal and switch to 'ZAI' (GLM 5.2). Now your agent uses GLM 5.2 as its brain.
  3. Run locally with Ollama: Download the free Ollama app, find the GLM 5.2 model, and plug it into tools like Claude Code, Codex App, OpenCore, or directly into Hermes Agent. You need a good local setup for this.

Julian recommends options 1 and 2 because they are easier and give you more tokens. With all three methods, you can start building right away.

04How to Use GLM 5.2 with the CLI and Hermes Agent

First, use the CLI method for coding. Open the dashboard, type something like 'build a 3D spaceship game', and GLM 5.2 writes the code and shows the game in your workspace. You can keep improving it later.

Julian built a full operating system with many apps this way.

Second, use Hermes agent for agentic workflows. You can voice‑activate Hermes by creating a custom voice agent. Then ask it to build things or schedule tasks.

All conversations and built projects are saved so you can preview them later.

Third, dive into the Kanban board. This is a native Hermes feature. You drop in a task, like 'write a blog post about GLM 5.2'.

The leader agent (using GLM 5.2) delegates to a writer agent, then an editor agent, and finally a judge agent. The judge checks quality and sends work back for improvement if needed. This creates a positive feedback loop that produces polished content.

05Results and Examples Shown in the Episode

Julian demonstrates several projects built with GLM 5.2 in one morning. He shows a full 3D game, a complete operating system with multiple apps, and various other tools. All projects are visually impressive and high quality.

For content creation, the Kanban board produced a published blog post and a video. The blog post is fully edited and beautifully drafted. The video went through iteration by the judge agent until it was ready to publish.

Julian says the quality is 'insane' and that this level of output would not have been possible a year ago.

He also points out that the coding plan gives so many tokens that he never ran out, even with heavy use. This makes GLM 5.2 cheaper than using an API for frontier models like Claude Opus.

06Tips, Limits and Mistakes to Avoid

One common mistake is thinking that cheap models are always worse. Julian shows that GLM 5.2 gives frontier results at a lower cost. Avoid being locked into one vendor.

If that vendor takes down a model, you lose your system. With GLM 5.2, you can run it yourself locally or through multiple platforms.

Another tip: don't wait for the next model. The people learning now will be way ahead later. Start using GLM 5.2 immediately.

For coding, use the CLI. For long‑horizon tasks like content creation, use the Kanban board with a team of agents. Always use the judge agent to check quality and iterate.

Finally, be careful when running locally. You need a very good setup to run GLM 5.2 on your own machine. If you don't have that, options 1 and 2 are better.

Also remember that Hermes agent with GLM 5.2 can be voice‑activated, which makes it even easier to use.

07Who GLM 5.2 Is For and Why It Matters

GLM 5.2 is for anyone who wants to build apps, games, websites, or automate content without technical skills. You can just type what you want and it builds it. It's also for people who want a cheaper alternative to expensive frontier APIs.

Business owners can use the agent operating system to automate daily tasks.

Julian's community of 3,600 business owners already use his agent operating system with GLM 5.2. They set up shared memory, shared context, and one dashboard to control all their agents. This means your agents understand your business and get smarter over time.

If you want to stay ahead with AI, learning GLM 5.2 now gives you a big advantage.