Chinese GLM-5.2 Model Rivals Claude and GPT at a Fraction of the Cost

Z.ai, a Beijing-based company listed on the Hong Kong stock exchange, has released GLM-5.2, a language model that scores just 1 percent behind Claude Opus 4.8 on long-horizon coding tasks and outperforms GPT-5.5. The biggest difference isn't the benchmark result, though, it's the price: GLM-5.2 costs five to seven times less than comparable services from OpenAI and Anthropic.

Pricing that hurts the competition

Z.ai prices a million input tokens at $1.40 and a million output tokens at $4.40. Those rates are nowhere near what top Western models charge, where similar work can cost several times more. The model runs with a context window of one million tokens, roughly 750,000 words held in working memory at once, which lets it handle large codebases and hours-long engineering tasks without losing context.

In Artificial Analysis's Intelligence Index v4.1 ranking, GLM-5.2 placed fourth in the world with a score of 51 points and topped the chart among open source models. In long-horizon coding projects it trails Opus 4.8 by only 1 percent while beating both GPT-5.5 and Opus 4.7. David Sacks, a former AI adviser in the Trump administration, said the model performs at a level close to OpenAI's and Anthropic's offerings, falling short only of their strongest variants.

Trained without Western chips

GLM-5.2 was trained on a cluster of roughly 100,000 Huawei Ascend 910B processors, with no Nvidia, AMD, or Intel chips involved at all. That shows Chinese companies can build models competitive with the global frontier despite US export restrictions on advanced chips. The model was released under an MIT license as open weights, so anyone can download it from Hugging Face, run it locally, and modify it freely, the only cost being their own computing infrastructure.

GLM-5.2's launch coincided with a period of tense tech rivalry between the US and China, including temporary export restrictions on models from American companies. Z.ai has already announced its next iteration, GLM-5.5, for August 2026, suggesting it plans to release new versions at the same pace as its Western rivals.

What it means for companies in Poland

For Polish companies looking for cheaper ways to deploy AI, especially in coding and process automation, GLM-5.2 is a concrete alternative that can be run locally without monthly licensing fees. The open MIT license means no regional restrictions and the ability to integrate it with in-house infrastructure, which matters for companies in sectors sensitive to data sovereignty.

At the same time, using a Chinese model raises questions about the origin of its training data, compliance with EU AI regulations, and trust in the infrastructure it was built on. Companies considering GLM-5.2 should check the terms under which input data is processed, especially if they use the version hosted by Z.ai rather than running the model in their own environment.

GLM-5.2's success also shows that price has become a major battleground in the language model market, alongside raw response quality. For companies with a limited AI budget, it's a sign worth testing alternatives to expensive subscriptions instead of automatically defaulting to the priciest option from the biggest providers.

Sources: What is GLM-5.2, the new Chinese AI model that's rivalling Anthropic (euronews.com), Chinese AI model catches up with OpenAI and Anthropic, costs a lot less (ithardware.pl), Chinese AI steps onto global stage as GLM-5.2 narrows frontier gap (news.cgtn.com).