Two major US artificial intelligence-powered coding tools released in the past week are suspected of being built on Chinese models, sparking debate about the ethics of commercialising open models without crediting their original developers.
Cognition AI’s latest SWE-1.5 model achieved “near state of the art” coding performance while also setting a record for generation speed, according to the San Francisco-based developer.
The company, valued at US$10.2 billion, said in a blog post that SWE-1.5 was built “on top of a leading open-source base model” without naming it, sparking speculation that it was Beijing-based Zhipu AI’s GLM series of foundational models after users asked the tool what model it was.
Zhipu AI, marketed internationally as Z.ai, said it believed SWE-1.5 used its latest flagship GLM-4.6 as the base model in a statement on Tuesday. Cognition AI did not respond to a request for comment.
Zhipu AI said it believed US company Cognition AI used its latest flagship GLM-4.6 as the base model. Photo: Shutterstock
Meanwhile, another US coding giant, Cursor, which recently tripled its valuation to US$9.9 billion in just six months, introduced a new tool called Composer, also with powerful coding capabilities and rapid generation speed.
However, users quickly noticed that the model generated reasoning traces in Chinese, leading some observers to conclude that its base model – the underlying model providing the raw intelligence for the tool – was also Chinese.