Key facts
- Chinese AI models narrowed the performance gap with US counterparts to 6% in June, down from 9% in May.
- This marks a record low in the performance gap.
- Zhipu AI's GLM-5.2 model achieved a top ranking in agentic coding.
- Two Chinese AI models were included in the top global rankings for large language models.
- Open-weight AI models, unlike open-source software, face significant business challenges due to high operational costs.
Chinese artificial intelligence models have significantly narrowed the performance gap with their US counterparts, reaching a record low of 6% in June, according to Bloomberg Intelligence. This development challenges the long-held notion of US technological supremacy in the AI sector.
Over the past year, the performance gap had typically ranged between 10% and 15%. However, the launch of Zhipu AI's GLM-5.2 model, which topped global rankings in agentic coding, and the inclusion of two Chinese models in LiveBench's global large language model performance list last month, signal a shift.
Despite the impressive technical achievements, the business model for "open-weight" AI, as exemplified by companies like Zhipu AI and MiniMax, is proving challenging. These companies have incurred substantial losses, with Zhipu AI losing nearly $500 million on $107 million in revenue last year, and MiniMax losing $250 million on $79 million in revenue. Their stock prices have also seen significant declines.
Analysts note that open-weight AI models differ fundamentally from open-source software. While software distribution costs are minimal, AI inference requires expensive hardware, electricity, and data center capacity. This high operational cost limits profitability, as seen when Moonshot AI had to halt new customer sign-ups for its Kimi K3 model due to insufficient computing power.
The economic model is further complicated because the trained numerical parameters of these open-weight models are often downloaded and run by third-party cloud providers or by the companies themselves, limiting direct revenue for the model creators. This contrasts with the success of companies like Red Hat in the open-source software space. Consequently, cloud giants like Alibaba have seen their stock prices rise while AI labs struggle.