Goldman Sachs elects three Chinese AI models; only one is open capital

Goldman Sachs has elected three Chinese artificial intelligence models as his preferred choices, but only one of them is traded in exchange. The highlight goes to GLM-5.2, an open-source model of Zhipu AI, which in several metrics rivals Anthropic’s Fable 5 — especially in the adoption by large companies and global SMEs. The selection was published in the bank's report and points to the growing Chinese role in the sector.

What are Goldman Sachs' favorite Chinese AI models?

The report cites three architectures: GLM-5.2 (from Zhipu AI), Qianwen 2.5 (from Alibaba Cloud) and DeepSeek-V3 (from DeepSeek). Of these, only Qianwen 2.5 is open capital, indirectly listed via Alibaba. The bank points out that GLM-5.2, being open-source and free, has attracted companies seeking to reduce licensing and customization costs — something that makes it especially competitive outside China.

Why is GLM-5.2 highlighted among Chinese AI models?

The GLM-5.2 surpasses important benchmarks in logical reasoning and long context understanding, comparing to Anthropic’s Fable 5. The differential is the cost: while the American model requires business subscription, GLM-5.2 can be downloaded and run locally by any developer. In Goldman Sachs internal tests, the Chinese model performed 12% higher in contract analysis tasks and 8% better in code generation — critical areas for business automation.

The impact of Chinese AI models for Brazilian developers and companies

For the startup and SME ecosystem in Brazil, the availability of a world-class open-source model such as GLM-5.2 reduces entry barriers. Companies can adapt it to local data without relying on expensive foreign APIs. In addition, the trend of smaller and more efficient models — as discussed in recent analysis of the AI race — accelerates adoption in sectors such as agribusiness, health and fintech.

How to invest in Chinese AI models?

The only asset directly exposed in the list is Alibaba (BABA), which offers Qianwen 2.5 via cloud. But Goldman Sachs recommends caution: competition with open models such as GLM-5.2 can push margins. For those who want indirect exposure, ETFs of Chinese technology or actions of local partners of Zhipu AI are alternatives. The full report, available on the bank's website, details that preference for open models is growing — by 2025, 34% of new AI deployments in global companies used Chinese open source technologies, according to CNBC itself — original source.

Frequently Asked Questions about Chinese AI models

Is GLM-5.2 really free for commercial use?

Yes, the model is distributed under MIT license, allowing commercial use without royalties, which makes it attractive for startups and SMEs.

What are the main risks of adopting Chinese AI models?

Data security issues and compliance with the General Data Protection Act (LGPD) may require adaptations, as well as possible geopolitical sanctions in regulated sectors.

Is DeepSeek-V3 available for Brazilian developers?

Yes, DeepSeek provides the model via API as well as open-source, but with a more restrictive license than GLM-5.2.