2025年第二季度中国人工智能现状亮点报告.docxVIP

StateofAI:China

ArtificialAnalysisQ22025

HighlightsReport

FullreportavailabletoPremiumAccesssubscribers

China’sleadingAIlabsarenowcloserthanevertoUSleaders,with

theleaddecreasingfrommorethanayeartolessthanthreemonths

CommentaryArtificialAnalysisIntelligenceIndexincorporates7evaluations:MMLU-Pro,GPQADiamond,HumanitysLastExam,

Commentary

ArtificialAnalysisIntelligenceIndexincorporates7evaluations:MMLU-Pro,GPQADiamond,HumanitysLastExam,LiveCodeBench,SciCode,AIME,MATH-500

Gemini2.5Pro,Google?TheperformancegapbetweenUSandChinesefrontiermodelssincethereleaseofChatGPTin2022has

Gemini2.5Pro,

Google

ArtificialAnalysisIntelligence

ArtificialAnalysisIntelligenceIndex

Grok3mini

reasoning(high),xAI

o3,

OpenAI

DeepSeekR1

(0528,May‘25)

o3-mini(high),OpenAI?DeepSeek’sopenweightsR1(May

o3-mini

(high),OpenAI

2025)modelleadstheChineseAI

o1,OpenAIQwen3235BA22B,Alibabao1-preview,OpenAIlabs,whileOpenAI’so3leadsmodelsreleasedbyUSAILabs

o1,

OpenAI

Qwen3235BA22B,Alibaba

o1-preview,

OpenAI

Seed-Thinking-v1.5,ByteDanceDeepSeekR1(Jan.25)GPT-4Turbo,

Seed-Thinking-v1.5,ByteDance

DeepSeekR1

(Jan.25)

GPT-4Turbo,

OpenAI

GPT-4o,OpenAIprimarilydriventheChinese

GPT-4o,OpenAI

GPT-4,OpenAIStep2,StepFunfrontier,whileadvancesin

GPT-4,

OpenAI

Step2,

StepFun

Yi-Large,01.AIQwen1.5Chat72B,AlibabaDeepSeekV3GPT-3.5Turbo,OpenAIQwenChat72B,

Yi-Large,

01.AI

Qwen1.5Chat

72B,Alibaba

DeepSeekV3

GPT-3.5Turbo,

OpenAI

QwenChat72B,Alibaba

GLM-4-Plus,

Qwen2.5Plus,

Alibaba

ZhipuAI

Qwen272B,AlibabaQwen1.5ChatQwenChat

Qwen272B,Alibaba

Qwen1.5Chat

`、、QwenChat7B,

`、、

QwenChat7B,Alibaba

DeepSeekLLM67BBaichuan4,

DeepSeek

LLM67B

Baichuan

?ArtificialAnalysis

Note:Someresultswereestimatedbasedoncompanyclaimsandcomparableresults

Source:ArtificialAnalysisIntelligenceIndex

Artificial

ArtificialAnalysisInte