A recent open-source AI model from China is making waves, with many drawing parallels to DeepSeek, the influential language model that left its mark on Silicon Valley.
Over the past week, developers, investors, and AI leaders have been lauding GLM-5.2, a model crafted by Z.ai, a company based in Beijing, especially for its capabilities in coding and agent-based tasks. The model boasts a 1 million-token context window, allowing it to process vast amounts of text simultaneously, rivaling established models like OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8. The biggest advantage? GLM-5.2 is available for free.
I put the model to the test across various tasks to assess its performance.
First up was drafting an email.
Right off the bat, I noticed that GLM-5.2 operated more slowly compared to premium alternatives and often faced capacity constraints. Whether it's worth the wait largely depends on one's patience.
After a few minutes, I tasked the model with crafting an outreach email for Business Insider to seek interviews with career coaches. The resulting email effectively covered all necessary points and closely resembled my typical writing style, aside from the lengthy processing time.
Next, I asked GLM-5.2 for recommendations on wet cat food suitable for a cat with a sensitive stomach.
Once I navigated through another delay, the AI provided a list of well-known brands, suggested a prescription option, and offered general advice for selecting food for sensitive cats. Although it took multiple attempts to bypass the capacity issues, I received a solid array of recommendations that aligned well with my veterinarian's advice over the years. However, GLM-5.2 currently lacks direct shopping links for the recommended products, which isn’t a major drawback, as a quick Google search can reveal where to buy them.
Moving on to the next task, I requested a weekend trip itinerary for two from Oakland to Monterey, California, focusing on hiking, scenic photography, antique shopping, dining options, and budget lodging.
The generated itinerary was comprehensive and thoughtful, featuring popular spots like Carmel and Moss Landing and factoring in potential traffic and reservations. However, there was a clear oversight regarding accommodations. Initially, it failed to suggest any budget hotels, and when prompted again, the options it provided, including the Super 8 by Wyndham Monterey, had prices that were unrealistic, listed around $100-$150 per night despite rates exceeding $300 at the time.
For the design task, I sought to create an advertisement for a fictional jewelry business featuring an Art Deco-style amethyst ring.
After more than 15 minutes stuck in a queue with GLM-5.2, I pivoted to the older 4.7 version. Surprisingly, this version processed my request in Chinese, resulting in both the reasoning behind its design choices and the final ad being presented in Chinese—an unexpected twist for those not familiar with the language.
Compounding the confusion, there were no options to download the output as a PDF or JPEG. When I requested a new version in English, the image vanished, and the HTML broke entirely.
Eventually, after some time, GLM-5.2 was back online. This newer model offered a more interactive design process, allowing me to choose styles and colors before generating the advertisement. While the output wasn't particularly stylish—it resembled a bar menu more than a high-end jewelry ad—it ultimately functioned well and allowed for a download.
For small businesses willing to invest the time in refining the model’s output, satisfactory results could likely emerge.
In conclusion, GLM-5.2, while not yet matching the sophistication or dependability of premium AI tools, presents a surprisingly competent option for a free, open-source model. Its current capacity limits and occasional delays can be bothersome, and certain features still show shortcomings. However, when it comes to everyday tasks like writing, research, shopping suggestions, and trip planning, GLM-5.2 often rivals the performance of far more costly competitors.
If Z.ai can enhance the model’s reliability and reduce wait times, GLM-5.2 may become an attractive alternative for those unwilling to shell out for premium AI subscriptions.
