AI Race: China vs USA - Models, APIs and Open Source

Company	Model Name	Capabilities	Target Model
Baidu (百度)	Wenxin Yiyan 4.5 Turbo / Wenxin X1 Turbo	Multimodal understanding, <128K context length, logical reasoning reaching GPT-4 level, visual + tool usage, multi-end deployment (UCloud)	GPT-4o / GPT-4
Alibaba (阿里巴巴)	Qwen·3 Series (Dense and Sparse), Qwen2.5-Omni, Multi-agent Qwen-Agent, Qwen-VL-Max	Open source model ecosystem, text/image/audio/video understanding, 32K–128K context, reasoning, coding, tool usage, multi-agent collaboration	LLaMA·3, GPT-4o
ByteDance (字节跳动)	Doubao (豆包) / Cici, Dreamina 3.0, Dreamina (Text-to-Image), Seaweed (视频)	Chat + text + image + video generation, SeedEdit image editing, 2M token text understanding / 100M token image understanding, HD image/video generation	Midjourney V6/V7
Tencent (腾讯)	Hunyuan·mix Series (混元)	Meeting notes / document collaboration / enterprise office services	❌ No clear overseas benchmark, leaning towards embedded deployment
Zhipu AI (智谱技术)	GLM-4 / ChatGLM3	Open-source representative in China, lightweight deployment, community ecology	LLaMA·2/3
Baichuan AI (百川智能)	Baichuan·2 (7B/13B/34B/192K), Baichuan·3 Series	Super long context (192K tokens), document understanding, reasoning	Gemini 1.5 Pro
DeepSeek (深度求索)	DeepSeek·V3 / DeepSeek·Coder	General reasoning, code generation, open-source activation	GPT-4 / CodeLlama
MiniMax	ABAB Series	Vertical customized services, multimodal landing product integration	❌ Product targeting only, no clear benchmark
iFLYTEK (讯飞)	Spark (星火 Spark·X)	Education scenario optimization + speech interaction + edge/local deployment	Gemini Nano (some difference in architecture targeting)
Moonshot AI (月之暗面)	Kimi-2 / Kimi-K2	Long text processing (up to 2000K context), strong QA, search enhancement, writing assistant, multi-document summarization	Claude 3.5·Opus (context + QA experience)

Related