Alibaba Introduces Tongyi Qianwen 2.0 and Industry AI Models

Chinese tech giant Alibaba Group Holding Ltd.’s cloud computing arm, Alibaba Cloud, has introduced an upgrade to its large language model (LLM) Tongyi Qianwen and launched artificial intelligence (AI) models for different industries amid fierce competition between tech firms.

The company said Tongyi Qianwen 2.0 is a significant step up from its predecessor, released in April.

The chatbot’s latest version is equipped with ‘hundreds of billions of’ parameters, referring to the AI model’s power, according to Alibaba Cloud. Such a measure makes it one of the world’s most powerful AI models.

Alibaba added that Tongyi Qianwen 2.0 features excellent capabilities in grasping complex queries, copywriting, thinking, learning, and avoiding hallucinations, which is the AI’s act of providing inaccurate information.

The Hangzhou-based company also announced eight AI models for applications in industries including entertainment, finance, healthcare, and legal. The industry-specific models provide tools for image creation, code writing, financial data analysis, and legal file searching.

The Race to Control China’s AI Market

The update coming six months after Tongyi Qianwen’s initial introduction demonstrated how fast tech firms are moving to access China’s promising and growing AI market.

In September, Chinese tech conglomerate Tencent Holdings Ltd. said the country was experiencing a ‘war of hundred models,’ being home to 130 LLMs or 40% of the world’s total, with industry-specific units connected to a main model.

The domestic rival of Alibaba said its Hunyuan AI has over 100 billion parameters and better processing of the local language than OpenAI Inc.’s GPT-4.

The US AI company’s GPT-3 model had 175 billion parameters in 2020, while Meta Platforms Inc.’s Llama 2 comprised 70 billion parameters for this year.

Alibaba Chairman Joe Tsai stated that the firm’s cloud computing unit currently oversees the operation of around half of the chatbots in China.

Alibaba’s text-to-video AI ModelScope now has 2,300 models and 2.7 million developers participating, according to Tsai.