As geopolitical and economic tensions grow between the US and China, another Chinese AI vendor has released a new major language model that offers the same features at a price that is apparently lower than Openai's LLM.
Moonshot AI, a startup supported by Chinese technology giant Alibaba, released its Kimi K2 model, the open source LLM on July 11th.
You K2
According to Moonshot AI, Kimi K2 has 32 billion active parameters and 1 trillion total parameters. There are two versions: Kimi-K2-Base and Kimi-K2-Instruct. The startup said the basic model is perfect for researchers who want fine-tuning and full control over custom applications.
The directed version, on the other hand, is a post-training model of generic chat and agent applications. The startup cited a benchmark that shows that the KIMI model is superior to DeepSeek-V3-0324 from another Chinese startup.
The release of the Kimi K2 evokes a similar response to the first release of Deepseek's R1 model earlier this year. Moonshot released Kimi K1.5 after Deepseek released the Deepseek-R1 in January. The vendor's open source model offers an alternative to unique models like Openai, raising questions about the inexpensiveness of AI technology. It will also be coming a few days after another Chinese vendor, Baidu, created an open source for the Ernie 4.5 model.
“Moonshot is positioned to attract a global developer base and enterprise user ecosystem with its open licensing, affordable API tiers and optional self-hosting options,” said Arun Chandrasekaran, analyst at Gartner.
He added that a mix of model expert architecture and chip innovation can connect lab research with real-world efficiency. For example, Moonshot provided an example of how Kimi K2 is being used to explore the pay of remote work.
Additionally, the Kimi K2 priced at $0.60 per million input tokens and $2.50 per million output tokens, down below the price of Openai's comparison model at an uncached price. On the other hand, Openai costs $2 per million, a $1 million output token.
Challenge and skepticism
However, model performance does not mean there are no challenges.
“As a model developed in China, overseas adoption could face regulatory or awareness challenges amid changing global technological dynamics,” Chandrasekaran said.
Additionally, Moonshot's K2 model may be an attempt to undermine the US model market, said David Nicholson, an analyst at Futurum Group.
“If a company outside the US sells its products to the market for less than the cost of producing these products, we call it dumping,” Nicholson said. “It's a strategy that controls a particular market. It's a market full of open and cheap. ”
He added that the big problem is the actual cost of models like the Kimi K2.
“China and the US are involved in the trade war,” Nicholson continued. “An announcement about large cost differences should be met with skepticism.”
Companies need to be aware of abandoning their data to China,” he said.
Meanwhile, Openai announced on July 12 that it would delay the release of its open source model indefinitely. CEO Sam Altman said the vendors need to perform more safety tests.
Additionally, Meta, known for its open source model, has continued its goal of advancing AI technology. The company recently acquired a voice generation AI vendor called Play AI. Play AI is known for using AI technology to create realistic, natural-sounding voices.
Esther Shittu is an Informa TechTarget News Writer and Podcast host that covers artificial intelligence software and systems.
