GLM-5.2 Review: How China’s Free AI Model Performed at Basic Tasks

A brand new open-source AI mannequin from China is drawing comparisons to DeepSeek, the LLM that shook Silicon Valley.

Builders, buyers, and AI executives have spent the previous week praising GLM-5.2, an open-source mannequin constructed by Beijing-based Z.ai, for coding and agentic AI duties. The corporate says it helps a 1 million-token context window — sufficient to course of lots of of 1000’s of phrases without delay — placing it in the identical league as OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8. In contrast to these fashions, GLM-5.2 is free.

I made a decision to check it throughout a spread of duties.

Here is how GLM 5.2 carried out at writing an e-mail, serving as a purchasing assistant, planning a visit, and making a poster.

Beginning easy: an e-mail

My first impression was that GLM-5.2 is noticeably slower than premium AI models and continuously runs into capability points. Whether or not it is price utilizing relies upon largely on how affected person you might be.

After ready a number of minutes, I requested it to write down an outreach e-mail for Enterprise Insider in search of interviews with career coaches.

The end result lined all of the necessities and intently matched the type of e-mail I might usually write myself. There was little to criticize past the wait time.

Selecting a product

Subsequent, I gave GLM-5.2 a activity I really care about: recommending moist cat meals for a cat with a delicate abdomen.

As soon as I received previous one other capability delay, the AI urged a number of well-known business manufacturers, a prescription possibility, and common recommendation for selecting meals for delicate cats.

Once more, it took just a few tries to get previous the mannequin’s capability subject. The AI ultimately offered a generic checklist of economic meals manufacturers for delicate and choosy cats, a prescription possibility, and a common algorithm for selecting cat meals, all of that are nice ideas for first-time pet house owners.

GLM, nevertheless, doesn’t present direct shopping links but for the merchandise it recommends.

As a lifelong caretaker of cats, I discovered the suggestions intently aligned with what my veterinarian has instructed me through the years. Whereas GLM does not present direct purchasing hyperlinks, I did not discover that to be a significant disadvantage — discovering the merchandise on Google takes just a few seconds.

For informational recommendation, I did not discover a lot distinction between GLM and costlier AI fashions.

Planning a visit

I additionally requested GLM to plan a weekend trip for 2 from Oakland to Monterey, California, together with mountaineering, scenic images spots, vintage purchasing, eating places, and a finances lodge.

The itinerary was considerate and detailed, recommending locations similar to Carmel and Moss Touchdown whereas accounting for site visitors and reservations.

The weak spot was lodging. Though I particularly requested a budget hotel, the mannequin initially skipped that half. After I requested once more, it urged a number of motels, however the costs weren’t reasonable. One suggestion, Tremendous 8 by Wyndham Monterey, was listed at roughly $100-$150 per evening, whereas present charges are effectively above $300.

Producing a design

The design check proved to be probably the most revealing.

I uploaded a photograph of an Artwork Deco-style amethyst ring and requested GLM to create an commercial for a fictional jewellery enterprise.

The most recent 5.2 mannequin spent greater than quarter-hour caught at capability, so I switched to the older 4.7 model. It unexpectedly processed my English immediate in Chinese language, producing each its reasoning and the completed poster in Chinese language.

I understand how to learn Chinese language, however a consumer who does not can be very confused at this level, as a result of this pondering course of resulted in a poster that is in Chinese language too, which I didn’t ask for.

To make issues worse, the end result doesn’t include a PDF or JPEG model I may obtain. After I requested it to regenerate the design in English, the picture disappeared, and the underlying HTML broke totally.

Finally, capability opened up on GLM-5.2. The newer mannequin took a special method, guiding me by an interactive design course of with color and style choices earlier than producing the ultimate poster. The design itself wasn’t particularly polished — it appeared extra like a bar menu than a luxurious jewellery commercial — however it functioned correctly and allowed me to obtain the end result.

A small business prepared to spend time iterating may in all probability produce one thing usable.

The underside line: it is price making an attempt

GLM-5.2 does not but match the polish or reliability of one of the best paid AI fashions. Capability limits are irritating, responses may be sluggish, and a few options — notably reside pricing and design era — nonetheless have apparent flaws.

Nonetheless, for a free, open-source model, it is surprisingly succesful. On on a regular basis duties like writing, analysis, purchasing recommendation, and journey planning, it typically delivers data that is corresponding to what you’d get from way more costly rivals.

If Z.ai can enhance reliability and scale back wait instances, GLM-5.2 may change into a compelling different for customers who do not need to pay for premium AI subscriptions.