Alibaba, Baidu, ByteDance heat up price war in China's AI market with steep LLM discounts, putting pressure on start-ups

Alibaba Cloud Intelligence, China's biggest cloud computing service provider, slashed the fees for using its generative artificial intelligence (AI) models by up to 97 per cent, a salvo in a price war that comes a week after ByteDance launched a rival service that costs less than most competitors.

The price cuts apply to nine of Alibaba's Qwen self-developed large language models (LLMs), also known as Tongyi Qianwen, said Liu Weiguang, president of public cloud business, at the firm's AI summit in Wuhan, capital of central Hubei province, on Tuesday.

"Today, we have lowered the threshold for enterprises to use LLMs better and faster, and for industries to create LLM-powered applications in the real world," Liu said.

Do you have questions about the biggest topics and trends from around the world? Get the answers with SCMP Knowledge, our new platform of curated content with explainers, FAQs, analyses and infographics brought to you by our award-winning team.

The discounts offered by Alibaba, owner of the South China Morning Post, come as Big Tech companies on the mainland race to draw users to their respective LLMs, the technology that underpins ChatGPT and other generative AI services. More than 200 LLMs have been introduced in China, according to government estimates.

Tongyi Qianwen, also known as Qwen, is a family of large language models developed by Alibaba. Photo: Shutterstock alt=Tongyi Qianwen, also known as Qwen, is a family of large language models developed by Alibaba. Photo: Shutterstock>

Hours after Alibaba Cloud's announcement, web search giant Baidu said its Ernie Speed and Ernie Lite LLMs, both launched earlier this year, are available free of charge with immediate effect.

The Beijing-based firm, which in November began charging 59.9 yuan a month for the use of its top-end Ernie Bot 4.0, is working to make its AI services more affordable, founder and chairman Robin Li Yanhong said in an earnings report last week.

That came a day after TikTok owner ByteDance introduced its self-developed Doubao LLM series, pricing its top-of-the-line model as low as 0.0008 yuan (US$0.0001) per 1,000-token prompt. In AI, a token is a fundamental unit of data that is processed by algorithms, with 1,000 tokens roughly equivalent to 750 English words.

For comparison, Alibaba Cloud's Qwen-max, the series' most powerful commercial model said to be capable of performance equivalent to OpenAI's GPT-4 Turbo, now costs 0.04 yuan per 1,000 input tokens after a 67 per cent discount.

Fees for Qwen-long, a commercial model designed to process long text, have been reduced by 97 per cent to 0.0005 yuan per 1,000 input tokens.

The company is also giving out 4 million free tokens per model for each new user until June 21, while offering a seven-day free trial for its open-source Qwen 1.5-110B and Qwen 1.5-32B.

Alibaba said earlier this month that its LLMs have been adopted by more than 90,000 corporate clients, including smartphone giant Xiaomi, a month after Baidu's Li said Ernie Bot had been used by 85,000 corporate clients.

The aggressive pricing tactics and expansive user bases of Chinese tech giants add pressure on top domestic AI start-ups trying to monetise their models. Zhipu AI currently charges 0.1 yuan per 1,000 tokens for its most capable GLM-4 LLM, while MiniMax's Abab 6.5 and Baichuan's Baichuan2-Turbo cost 0.03 yuan and 0.008 yuan respectively.

This article originally appeared in the South China Morning Post (SCMP), the most authoritative voice reporting on China and Asia for more than a century. For more SCMP stories, please explore the SCMP app or visit the SCMP's Facebook and Twitter pages. Copyright © 2024 South China Morning Post Publishers Ltd. All rights reserved.

Copyright (c) 2024. South China Morning Post Publishers Ltd. All rights reserved.