(Yicai) May 9 -- Alibaba Group Holding's cloud computing arm released a new version of its large language model Tongyi Qianwen, boasting that its performance score on a third-party review platform has caught up with OpenAI's GPT-4 Turbo.

Tongyi Qianwen 2.5 scored 50 on OpenCompass, the Shanghai Artificial Intelligence Laboratory's LLM review platform, the same as GPT-4 Turbo, Alibaba Cloud announced today. It is the first China-made LLM to achieve such a score, it added.

Tongyi Qianwen 2.5 has improved its comprehension by 9 percent, logical reasoning by 16 percent, command following by 19 percent, and coding capabilities by 10 percent compared with its 2.1 version, the Hangzhou-based firm noted. It has also surpassed GPT-4's text comprehension, text generation, knowledge quiz, and life advice functions in Chinese, it pointed out.

Released in April last year, Tongyi Qianwen has served more than 2.2 million enterprise customers via Alibaba's office application DingTalk, Alibaba Cloud said. Its open-source model has been downloaded over 7 million times and applied to various industries, including education, healthcare, food and beverage, gaming, and culture and tourism, it added.

ChatGPT's developer OpenAI launched GPT-4 Turbo last November. The turbo version supports a conversation length of 128 kilobytes, compared with GPT-4's 8 kb, or about 6,000 words. It also adopts a knowledge base that was updated last April.

With artificial intelligence's continued iteration, the reference for Chinese tech companies in developing LLMs has shifted from the original ChatGPT to upgraded versions. Shanghai-based SenseTime Group released SenseNova 5.0 last month, claiming that its comprehensive capability can take on GPT-4 Turbo and reason about 200 kb of text.

Editors: Dou Shicong, Martin Kadiev