Battle price at Alibaba: AI applications become cheaper

Alibaba is making its large language models up to 85% cheaper. This shows that the price war in AI is going – downwards.

listen Print view

(Image: Peshkova / Shutterstock.com)

3 min. read

Qwen-VL, Alibaba's vision language model, will be offered 85 percent cheaper in the future. This information comes from a WeChat post by Alibaba Cloud, writes CNBC. It indicates that the price war for AI models has begun. However, it is not that better models are becoming more and more expensive, but rather that the models are becoming cheaper and cheaper due to the enormous competition and lack of differences. This is evidently happening in China first, but is not sparing Western markets either. Alibaba had already repeatedly lowered the prices of other models last year.

In China, there are comparable generative AI models and applications from popular tech companies such as Alibaba, Tencent, Baidu, JD, Huawei and ByteDance. In addition, there is Deepseek, an AI-specialized company that has not only trained its AI model of the same name at low cost, but also offers open source.

Alibaba's Qwen is not a product for end consumers, it is an AI model that can be used by companies. While the prices for a subscription to ChatGPT remain stable at around 20 euros per month, the costs for access to AI models have actually changed for OpenAI, Google and the like. On the one hand, usage costs in the B2B sector have fallen, while on the other, providers have launched smaller, more cost-effective models on the market that are becoming increasingly powerful.

Videos by heise

Due to the high costs of training AI models, all providers are also working on methods to make this cheaper. This includes model distillation, for example, in which the knowledge of a large model is transferred to a smaller one.

The cost pressure is also fueled by numerous open source or freely available models. These include Meta's Llama AI family, which can be used free of charge. However, Mark Zuckerberg also clearly states that this is only possible because Meta can cross-subsidize AI development. The company earns money from advertising on social networks, among other things.

It is known that OpenAI is also under enormous pressure because the company has been dependent on donors up to now. Accordingly, the company's website also states that investments should be recorded as donations. However, OpenAI is planning to change its own corporate structure from a non-profit company to a profit-oriented company – so that investors can also get more money. To accomplish this, however, OpenAI would first have to earn more money. The plan is apparently for models such as o1 and o3 to generate more revenue. OpenAI emphasizes their strong reasoning abilities, meaning that they are supposed to be particularly good at logical thinking. However, there are also doubts about this.

(emw)

Don't miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.