01What is GLM 5.2 vs Claude Code?

GLM 5.2 vs Claude Code is a head-to-head comparison between China's new open-source AI model and Anthropic's powerful coding tool. In this episode, Julian Goldie tests them side by side on real tasks like building games and websites.

GLM 5.2 is fully open source and MIT licensed, so anyone can run it for free. Claude Code is closed, so you must pay for API access to use it in your own agents. That difference matters a lot.

02Main Features of GLM 5.2

GLM 5.2 has several standout features that make it different from Claude Code:

  • Open source – available on Hugging Face and runs on Ollama for free.
  • Design leader – scored number one on Design Arena, beating Claude Fable 5.
  • Agent friendly – plugs into Hermes or other AI agents using OAuth, no API needed.
  • Cheap coding plan – costs less than Claude for coding tasks.
  • Customizable outputs – in tests, GLM 5.2 gave more options like zoom, drag, and theme changes.

These features make GLM 5.2 a practical choice if you want control and low cost.

03How to Set Up GLM 5.2

Getting GLM 5.2 running is simple. Here are the steps Julian showed:

  1. Go to Hugging Face and find the GLM 5.2 model page.
  2. Download the model or use the hosted version.
  3. Install Ollama on your machine if you want to run it locally.
  4. Pull GLM 5.2 into Ollama using the command line.
  5. Connect it to your agent like Hermes or OpenClaw using OAuth login.
  6. Optionally, set it as the brain inside Claude Code or Codex.

That's it. You're ready to build and test.

04How to Use GLM 5.2 in Tests

In the episode, Julian used GLM 5.2 to build several mini-games. He compared it directly with Claude Opus 4.8 and Kimi K 2.7.

Tests included a first-person raycaster engine, a flying neon city, a racing game, a solar system orbit, a voxel runner, a liquid-in-a-bowl simulation, and a landing page. For each test, he ran the same prompt in all three models and judged the result.

GLM 5.2 won the voxel runner and liquid bowl tests with more fun, interactive visuals. It also built a more animated landing page than Claude. But Claude Opus 4.8 won the raycaster, neon city, racing game, and the second solar-system-like test.

The key takeaway: GLM 5.2 works well for creative, design-heavy tasks, while Claude handles smooth playability better.

05Benchmarks and Test Results

Julian also looked at public benchmarks. On SWE-Bench Pro, GLM 5.2 scored 62.1, while Claude Opus 4.8 scored 69.2. On Terminal Bench, GLM 5.2 got 69 and Opus 4.8 got 48, meaning GLM beat Claude on that one.

In the agent arena, Claude Fable 5 and Opus 4.8 are top, but you can't use them with your agents without paying for API. GLM 5.2 supports OAuth, so you can log into agents directly.

In the hands-on tests, the final score was a draw: both GLM 5.2 and Opus 4.8 won three tests each. Kimi K 2.7 didn't win any.

Overall, Claude Opus 4.8 is still the most powerful and smoothest, but GLM 5.2 is very close and cheaper.

06Tips, Limits, and Mistakes to Avoid

There are some important limits to know before you switch to GLM 5.2. Claude Opus 4.8 still wins on smoothness and hard coding tasks. GLM 5.2 can feel buggy in first-person games, and controls are hard sometimes.

Don't expect GLM 5.2 to beat Claude on every test. Results vary depending on the prompt, as Julian saw when he ran different tests on different days.

Also, avoid the mistake of paying for Claude API if you only need agent work. GLM 5.2 is free and works with OAuth, so use that instead.

If you want the best of both worlds, you can even plug GLM 5.2 into Claude Code as the brain. That's a clever workaround.

07Who Should Use GLM 5.2

GLM 5.2 is for anyone who wants a powerful AI without the high cost. If you build with agents, run a small business, or just like experimenting, this model is a great pick.

It's especially good if you care about design. GLM 5.2 is number one on Design Arena, so it beats even Claude Fable 5 on design tasks.

Open source means you can run it anywhere, even offline. That freedom is something Claude can't offer.

If you do serious coding all day, stick with Claude Opus for now. But for everything else, GLM 5.2 is a smart, affordable choice.

08Why GLM 5.2 Matters

GLM 5.2 matters because it shows open-source AI can compete with the big paid models. It doesn't win every test, but it wins on cost, flexibility, and design.

In the episode, Julian used it inside his own agent operating system. He said everything you build with GLM 5.2 gets saved, and you can switch agents easily.

He also mentioned that Claude Fable 5 got shut down this week, but GLM 5.2 keeps running because it's open source. That's a huge advantage.

For people who want control and low cost, GLM 5.2 is a real alternative. It's not the best at everything, but it's the best for many people.