GLM-5.2: The Chinese AI Challenging Silicon Valley Giants

GLM-5.2: A Chinese Alternative That Intrigues Silicon Valley

The artificial intelligence model GLM-5.2, developed by the Beijing-based company Z.ai, is starting to make waves in Silicon Valley. This model stands out due to its open-source approach and its free availability, attracting the attention of developers, investors, and leaders in the AI field. With an impressive context of 1 million tokens, GLM-5.2 positions itself as a potential competitor to well-established models such as GPT-5.5 from OpenAI and Claude Opus 4.8 from Anthropic. While its coding and AI agent capabilities are praised, it is its free accessibility that makes it particularly appealing.

Email Writing Test

To evaluate the performance of GLM-5.2, I undertook testing it on several tasks. The first was to draft a professional email to Business Insider requesting interviews with career coaches. From the outset, it became clear that GLM-5.2 is slower than its paid counterparts, and it often faces capacity limitations. After a prolonged wait, the model produced an email that covered all the essential points, with a style similar to what I would have written myself. However, the waiting time remains a significant drawback.

Product Recommendations

Next, I asked GLM-5.2 to recommend wet food for a cat with a sensitive stomach. After overcoming some capacity-related hurdles, the AI suggested several well-known commercial brands, as well as a prescription option and advice on choosing food for sensitive cats. Although the model did not provide direct purchase links, the recommendations aligned with my veterinarian's advice, which reassured me of their relevance. Finding the mentioned products through a quick Google search posed no major issues.

Trip Planning

To test its planning capabilities, I asked GLM to prepare a weekend itinerary for two people traveling from Oakland to Monterey, California. The proposed itinerary was well thought out, including suggestions for hikes, photo spots, antique shopping, restaurants, and accommodations. Specific destinations such as Carmel and Moss Landing were recommended, taking into account traffic and necessary reservations. However, the model initially failed to suggest an economical hotel, as requested. After a follow-up, it suggested motels, but the prices indicated were well below actual rates, particularly for the Super 8 by Wyndham Monterey, listed at around $100-150 per night, while current rates exceed $300.

Design Creation

The graphic design test was particularly revealing of the current limitations of GLM-5.2. By uploading a photo of an Art Deco-style amethyst ring, I asked the model to create an advertisement for a fictitious jewelry company. Version 5.2 took over 15 minutes to respond due to capacity issues, forcing me to revert to the earlier version 4.7. The result, generated in Chinese, showed some confusion for users not fluent in the language. Additionally, the lack of downloadable formats like PDF or JPEG complicated the use of the produced design. Ultimately, the capacity freed up on GLM-5.2, and the model adopted a different approach, guiding the user through an interactive design process with style and color options before generating the final poster. The design, while functional, resembled more of a bar menu than a luxury jewelry advertisement, but it finally allowed for the result to be downloaded.

Conclusion: Untapped Potential

Although GLM-5.2 cannot yet compete with the polish and reliability of paid AI models, it offers surprising performance for a free and open-source model. Its capabilities in everyday tasks such as writing, information retrieval, product recommendations, and trip planning are often comparable to those of more expensive models. If Z.ai can improve reliability and reduce waiting times, GLM-5.2 could become a serious alternative for those looking to avoid the subscription fees of premium AIs.