2025年第一季度中国人工智能发展状况报告

Artificial AnalysisQ1 2025 Artificial Analysisis a leading and independent AI benchmarkingandinsights provider.We supportengineers and companies to understand AIcapabilities and make critical decisions about their AI strategy. Ourdata, insights and publications are grounded in our comprehensivebenchmarking of AI technologies and use cases. This includes everythingfrom hourly performance testing of language model APIs to millions ofvotes in our crowd-sourcedarenas. Our public website,artificialanalysis.ai,is widely referenced by companiesleading innovation in AI. To discuss this report, our publications, or ourservices, please get in touch atcontact@artificialanalysis.ai. FRONTIER LANGUAGE MODELS BY ORIGIN Chinese AI labs have progressively caught up to US AI labs; models from Chinese labsare now approaching o1-level intelligence with the release ofDeepSeek’sR1 model Key Trends Closing the gap:The final months of 2024 haveseen the emergence of thenumerous highlyperformant models from top Chinese AI labs. Thishas resulted in the delta between the level ofintelligence offered by models from Chinese AIlabs and US AI labs closing. Several Chinesemodels are now competitive with models from thetop US labs. Reasoning models quickly becomingcommonplace:Reasoning models (that “think”before answering) were first introduced by OpenAIin 3Q24. Within months, Chinese competitors, ledby DeepSeek, have largely replicated theintelligence of o1. Several AI labs in China nowhave a frontier-level reasoning model. Open models close in on thefrontier labs:Openweights models, led by those from DeepSeek andAlibaba, have approached o1level intelligence. LEADING USFRONTIER LANGUAGE MODELS Since the launch of OpenAI’s GPT-4 in early 2023, leading US AI labs have scrambledto catch up to OpenAI Leading US AI Labs Frontier Language Model Intelligence, Over Time1Key Trends Competing labs catch up to OpenAI’s GPT-4:OpenAI started the language model race inNovember 2022 with the launch of GPT-3.5inChatGPT; leading US labshave largely caught upwith frontier models from OpenAI. Big Tech closes in on thefrontier labs:Modelsfrom Google and Meta are rapidly closing in onfrontier models, with Gemini 2.0 Flash exceedingClaude 3.5 Sonnet and GPT 4o capabilities. Sparks of intelligence beyond GPT-4:The finalmonths of 2024 have seen the emergence of thefirst major intelligence leaps beyond GPT-4, led byOpenAI’s o3.Topics including reasoning models,data quality and new reinforcement learningtechniques have joined pre-training computescaling as dominant levers for improving models. LEADING CHINESEFRONTIER LANGUAGE MODELS Leading Chinese AI labs DeepSeek and Alibaba have steadily released new models,with DeepSeek taking the lead from Alibaba in late 2024 Key Trends Rapid improvements in intelligence:WhileChinese AI labs joined the AI race later, they largelyclosed the intelligence gap with frontier US modelsin 2024. When OpenAI launched o1, Chinese labsproduceda similarly performant model withinmonths (DeepSeek’sR1). Leading with open weights models:Chinese AIlabs, including Alibaba, DeepSeek and Tencent,have released open weights frontier models that arecompetitive with the leading models globally. Potential leader in 2025:Early 2025 saw Chinese AIlabs, including Alibaba, DeepSeek,MoonShot,Tencent,Zhipu, andBaichuanprolifically releasingfrontier reasoning models. The release velocity andcadence suggest that Chinese AI labs are no longerlaggards in 2025. LANGUAGE MODEL COUNTRY OF ORIGIN While the US maintains an overall lead in the intelligence frontier, China is no longerfar behind. Few other countries have demonstrated frontier-class training LANGUAGE MODEL COUNTRY OF ORIGIN As of early 2025, several Chinese AI labs have demonstrated or claimed frontier-levelintelligence, with seven releasing models featuring reasoning capabilities The Language ModelFrontier: Models by Chinese AI Labs Artificial Analysis Intelligence Index, Leading Models (Early 2025), Non-exhaustive CHINA AI LABS OVERVIEW: BIG TECH The leading Chinese Big Tech firms are actively competing in the AI race and havereleased AI language models as well as models across other modalities Non-Exhaustive Open Weights LLM Other Firms with AI Ambitions Kunlun TechSHE: 300418 (Mkt Cap: $6B)1 Beijing-based internet group with >300m MAUs;owner of the Opera browser. Launched theSkyWorkseries of models and AI accelerators 360 Security (Qihoo 360)SHA: 601360 (Mkt Cap: $11B)1 China's largest provider of Internet and mobilesecurity products. Launched theZhinaoseries ofmodels under the 360 AI brand iFlytekSHE: 002230 (Mkt Cap: $16B) 1 Leading voice AI company in China with >14,000employees. Launched the Spark series of models Meituan HKG: 3690 (Mkt Cap: $115B)1 China’s leading shopping platform with >600mDAUs. Cofounder WangHuiwenreturned to leadAI efforts. Investor in multiple frontier AI labs Xiaomi China’s leading consumer electronics brand.HKG: 1810 (Mkt