Over the past week, developers, investors, and AI executives have been praising GLM-5.2, an open source model for coding and agent AI tasks built by Beijing-based Z.ai. The company says it supports a 1 million token context window, enough to process hundreds of thousands of words at a time, and is on par with OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8. Unlike these models, GLM-5.2 is free.
I decided to test it across a variety of tasks.
Here’s how GLM 5.2 performs as it composes emails, acts as a shopping assistant, plans trips, and creates posters.
Get started easily: email
My first impression was that the GLM-5.2 was noticeably slower than the premium AI model and had frequent capacity issues. Whether it’s worth using depends largely on how patient you are.
After waiting a few minutes, I asked them to write an outreach email for Business Insider asking for an interview with a career coach.
The results covered all the important points and were pretty consistent with the style of emails I usually write myself. There was little to criticize other than the wait time.
Product selection
Next, I gave GLM-5.2 a task that I was actually interested in. Wet cat food is recommended for cats with sensitive stomachs.
Once I got over the capacity lag again, the AI suggested several well-known over-the-counter brands, formula options, and general advice for choosing food for sensitive cats.
Again, it took several tries to resolve the model capacity issue. The AI ultimately provided a general list of commercial food brands for sensitive and picky cats, prescription options, and general rules for choosing cat food. These are all great tips for first-time pet owners.
Having cared for cats all my life, I’ve found that the recommendations are pretty much in line with what veterinarians have been telling me for years. GLM doesn’t offer direct shopping links, but I didn’t find that to be a big drawback since it only takes a few seconds to find a product on Google.
As a tip, there was no significant difference between GLM and the more expensive AI model.
plan your trip
I also asked GLM to plan a weekend trip for two from Oakland to Monterey, California. This includes hiking, beautiful photo spots, antique shopping, restaurants, and affordable hotels.
The itinerary was thoughtful and detailed, recommending destinations like Carmel and Moss Landing while taking into account traffic and reservations.
The weak point was accommodation. I specifically requested a business hotel, but the model initially skipped that part. I asked again and they suggested a few motels, but the prices weren’t realistic. One of our recommendations, the Super 8 by Wyndham Monterey, was listed at around $100-$150 per night, but current rates are well over $300.
Generate the design
Design testing turned out to be the most revealing.
I uploaded a photo of an art deco amethyst ring and asked GLM to create an ad for my fictional jewelry business.
The latest 5.2 model ran out of space for more than 15 minutes, so I switched to the older 4.7 version. Unexpectedly, it processed my English prompt in Chinese and produced both its reasoning and the completed poster in Chinese.
I know how to read Chinese, but users who don’t will be very confused at this point. Because this thought process also resulted in a poster in Chinese, which I didn’t ask for.
To make matters worse, the results don’t come with a downloadable PDF or JPEG version. When I asked them to regenerate the design in English, the images disappeared and the underlying HTML was completely broken.
Finally, GLM-5.2 freed up space. The new model took a different approach, guiding us through an interactive design process with style and color options before producing the final poster. The design itself wasn’t particularly elegant, looking more like a bar menu than an advertisement for fine jewelry, but it worked properly and I was able to download the results.
If you’re a small business that takes the time to iterate, you’ll probably be able to produce something usable.
Conclusion: worth a try
GLM-5.2 still cannot match the sophistication or reliability of the best paid AI models. Capacity limits can be frustrating, responses can be slow, and some features still have obvious flaws, especially live pricing and design generation.
Still, it’s surprisingly capable for a free, open-source model. For everyday tasks like writing, research, shopping advice, and travel planning, they often provide information on par with what you get from more expensive competitors.
If Z.ai can improve reliability and reduce latency, GLM-5.2 could become an attractive alternative for users who don’t want to pay for a premium AI subscription.
