Two new Chinese artificial intelligence models have broken into a top 10 leaderboard for both open- and closed-source models, highlighting China’s progress in closing the gap with leading US developers.
Alibaba Group Holding’s 1 trillion-parameter Qwen3-max-preview model debuted in sixth place in the latest “text arena” ranking by LMArena – an AI model evaluation platform started by University of California, Berkeley researchers – making it the top Chinese model.
Meanwhile, Alibaba-backed start-up Moonshot AI’s updated Kimi-K2 model tied for eighth place with seven other models, including DeepSeek R1 and xAI’s Grok 4, reinforcing its status as one of the premier open-source models globally.
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.
Qwen3-max-preview – Alibaba’s largest AI model to date – and Kimi-K2-0905 were both released on Friday. Alibaba owns the South China Morning Post.
The achievement stands in stark contrast to a year ago, when no Chinese models were represented in LMArena’s top 10 leaderboards, which is based on user feedback. Users submit prompts to two anonymous models and vote for the one that provides the better answer. The models’ identities are then revealed.
Unlike static benchmarks, LMArena’s dynamic “arena” format pits models against each other, resulting in rankings that evolve based on user preferences and the latest model updates.
With a score of 1,428 on LMArena’s text leaderboard, which assesses models based on text generation, Qwen3-max-preview ranks just behind leading models from US giants OpenAI, Anthropic and Google DeepMind.
In contrast to other models in the Qwen series, Alibaba has not open-sourced Qwen3-max-preview, and its full launch date has yet to be announced.
Reception for Kimi-K2-0905 has also been positive, with users praising its affordability and newly expanded context window, both crucial for demanding coding tasks. Open-sourced since Friday, it has become one of the top-trending models on AI developer platform Hugging Face, with more than 4,500 downloads.
According to the overall intelligence index of Artificial Analysis, a leading AI benchmark consultancy, Kimi-K2-0905 improved by just two points compared with the original K2 model launched in July, and trailed behind leading closed-source US models, as well as Chinese open-source models like Qwen3 235B 2507 and DeepSeek R1 0528.
However, the updated model showed substantial enhancements in agentic capabilities, including advanced coding and customer service, the consultancy said.
AI has emerged as one of China’s fastest-growing industries. The number of Chinese AI companies surged to more than 5,000 currently from 1,454 in 2020 – corresponding to about one new company every 11 hours, according to data released by the Ministry of Industry and Information Technology on Monday.
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 © 2025 South China Morning Post Publishers Ltd. All rights reserved.
Copyright (c) 2025. South China Morning Post Publishers Ltd. All rights reserved.