01What is GLM 5.2 and How It Works

GLM 5.2 is a new coding AI model from the Chinese company Z.ai (also called Jiabu ZAI). It is designed to write code and create interactive projects like games and websites. In this episode, the model was tested head-to-head against Claude Opus 4.8 and Qwen 3.7 Max.

The models were given five identical tasks: a voxel runner game, a solar system orbit map, a liquid-in-a-bowl animation, a landing page, and an arcade game. Each had one shot, no second chances. The results showed that GLM 5.2 won most tests, even though it had no official benchmark scores.

However, when used inside an AI agent, it was slower and less useful than Qwen.

02GLM 5.2's Main Features in Real Tests

GLM 5.2 shows several strengths when used for coding projects. It creates fun, interactive games with smooth animations. It also builds complete landing pages with nice visual effects.

  • Fun and engaging games (voxel runner, arcade game).
  • Interactive animations like liquid effects that you can change.
  • Detailed landing pages with multiple sections and animations.
  • Produces finished projects that are more interesting than competitors.

In the episode, GLM 5.2 also powered an open-world game and a Skyrim-style RPG. These were created inside the agent operating system, showing its ability to handle complex coding. Yet, it lacked official benchmark numbers, so its theoretical performance is unknown.

03How to Set Up and Use GLM 5.2

To start using GLM 5.2, you likely need a coding plan from a service like Hermes, where you can plug it into your agents. The episode shows that you can create separate profiles for GLM 5.2, Qwen 3.7 Max, and Kimi K 2.7. This lets you switch models easily.

  1. Get a coding plan that supports multiple AI models (e.g., Hermes).
  2. Look for the option to add a GLM 5.2 profile.
  3. Plug GLM 5.2 into your existing agent workspace.
  4. Choose GLM 5.2 when you want to code directly in the CLI.
  5. Use the agent operating system to preview your projects live and see agents work together.

You can also use GLM 5.2 to create complete projects by giving a single prompt, as shown with the video agents and games.

04How to Use GLM 5.2 for Real Coding Tasks

The episode shows several ways to use GLM 5.2, mainly for coding and creative builds. For example, you can ask it to create a game, and it will produce a playable result with animations. In the voxel runner test, the game was fun and had good graphics.

For a landing page, GLM 5.2 generated a full site with scrolling sections and nice animations. You can then preview it in the agent operating system's workspace and even run it live.

When using GLM 5.2 inside an agent like Hermes, you can ask it to review your notes and do tasks like SEO keyword research. However, in the test, it gave a brief answer, unlike Qwen, which provided detailed suggestions. So for agent tasks, you might prefer another model.

05Test Results: GLM 5.2 vs Claude and Qwen

In the episode, eight tests were shown. GLM 5.2 won four: the voxel runner, liquid in a bowl, landing page, and arcade game. Claude Opus 4.8 won only the solar system orbit map.

Qwen 3.7 Max won none of the five main tests, though it performed better in agent-based tasks.

Benchmarks told a different story. Qwen 3.7 Max reported 80.4% on SWE-Bench Verified, making it look the strongest. But in real use, it was last.

Claude Opus 4.8 was too new to be included in Qwen's benchmark comparisons. This shows that benchmarks don't always match real-world results.

When used in an AI agent, Qwen 3.7 Max was faster and gave better SEO keyword research than GLM 5.2. So the best model depends on the task.

06Tips, Limits and Mistakes to Avoid

When using GLM 5.2, keep in mind a few things from the episode. First, it may be slower than Qwen when used in an AI agent, so be patient. Second, it gives brief answers for non-coding tasks like SEO, which might not be helpful.

  • Don't rely only on benchmarks; real tests matter.
  • Choose GLM 5.2 for creative coding, but not for agent-based research.
  • Use the agent operating system to combine multiple models.
  • Check response speed—Qwen was faster in agents.

Another mistake is using only one model. The episode recommends having all three (GLM, Qwen, Claude) in one place to leverage each model's strengths. For orchestrating agents, Claude was better; for cool projects, GLM was best.

07Who Should Use GLM 5.2 and Why

GLM 5.2 is ideal for developers and creators who want to build fun, interactive projects. If you enjoy coding games or designing landing pages, you'll appreciate its creativity and polish. It's also a strong choice for CLI coding tasks.

For business users who rely on AI agents for analysis or SEO, you might prefer Qwen 3.7 Max due to its speed and detailed answers. However, adding GLM 5.2 to your toolkit gives you flexibility—you can switch based on the task.

The episode also shows that using multiple AI models together can be powerful. By having all three in one dashboard, you get the best of each. This approach is especially valuable for teams that need to produce varied content, from websites to videos.