Skip to content

TheRouter-ai/awesome-chinese-ai-models

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

71 Commits
.github
.github
 
 
agent
agent
 
 
collections
collections
 
 
data
data
 
 
docs
docs
 
 
newsletters
newsletters
 
 
schema
schema
 
 
scripts
scripts
 
 
tests
tests
 
 
.gitignore
.gitignore
 
 
CONTRIBUTING.md
CONTRIBUTING.md
 
 
README.md
README.md
 
 
requirements.txt
requirements.txt
 
 

Repository files navigation

Awesome Chinese AI Models

A global guide to China's AI model ecosystem. Discover, compare, and access leading AI models developed in China.

Chinese AI labs — DeepSeek, Qwen, GLM, Kimi, MiniMax, ERNIE, Hunyuan — now ship some of the most capable and cost-efficient models available, and many are open-weight. But for developers outside China, the ecosystem is hard to navigate: scattered docs, unclear pricing, uncertain global availability.

This repository is a curated, continuously updated entry point. It covers:

  • Model comparison — capabilities, context windows, open weights
  • How to access — official APIs, routing gateways, and self-hosting
  • Best-for guidance — which model fits which workload
  • Global availability — where these models can be reached, and how to pay
  • Benchmarks — aggregated from established leaderboards
  • Daily updates — releases, API changes, and ecosystem news

中文：一个持续更新的中国 AI 模型索引——对比、接入方式、全球可用性、benchmark 和每日生态新闻。 Maintained by TheRouter.ai for the global developer community.

Contents

Model Comparison

A quick map of the major Chinese AI model families. DeepSeek is a popular low-cost alternative for coding workloads; Qwen ships strong multilingual open-weight models; GLM and Kimi focus on reasoning and long context.

Model Best For Context Vision Reasoning API Open Weights
DeepSeek Low-cost coding and reasoning 128K — ✅ ✅ ✅
Qwen Multilingual and open-weight deployment 256K ✅ ✅ ✅ ✅
GLM Chinese reasoning and agent workflows 128K ✅ ✅ ✅ ✅
Kimi Long-context document workloads 256K — ✅ ✅ —
MiniMax Long-context, audio, and video generation 1M ✅ ✅ ✅ ✅
ERNIE Enterprise deployment on Baidu Cloud 128K ✅ ✅ ✅ ✅
Hunyuan Multimodal generation including 3D world models 256K ✅ ✅ ✅ ✅
Doubao Multimodal agents on Volcengine Ark 256K ✅ ✅ ✅ —

Context windows and capabilities are best-effort and change often. Always verify with the official provider.

How to Access

There are several ways to use Chinese AI models. This section stays vendor-neutral — official APIs, third-party aggregators, and self-hosting are listed side by side so you can choose what fits your project.

  • Official API — Call the provider's own API directly.
    • Best for: Production workloads that need first-party SLAs, billing, and the newest model versions.
    • Each provider has its own endpoint, pricing, and regional availability. Most expose an OpenAI-compatible mode.
  • TheRouter — An OpenAI-compatible routing gateway with fallback and cost controls.
    • Best for: Teams that want one endpoint across multiple Chinese and global models, with routing and accounting.
    • Maintained by TheRouter.ai, the maintainer of this repository. Listed as one option among several.
  • OpenRouter — A third-party aggregator that exposes many models behind one API.
    • Best for: Quick experimentation across providers without separate accounts.
    • Independent third party. Coverage and pricing of Chinese models varies.
  • Self-host — Run open-weight models on your own infrastructure.
    • Best for: Data residency requirements, offline use, and full control over inference.
    • Only available for models that release open weights. Inference stacks include vLLM, SGLang, and llama.cpp.

Best For

Developers new to Chinese AI models often ask which one to start with. This mapping is a starting point, not a ranking — benchmark on your own workload before committing.

  • DeepSeek
    • Low-cost coding workloads
    • A cost-efficient alternative for reasoning tasks
  • Qwen
    • Multilingual applications
    • Open-weight self-hosting and fine-tuning
    • Terminal coding agents (Qwen-Code)
  • GLM
    • Chinese-language reasoning
    • Agent and tool-use workflows
  • Kimi
    • Long-context document analysis
    • Retrieval-heavy workflows
  • MiniMax
    • Very long context windows
    • Audio and video generation
  • ERNIE
    • Enterprise integration on Baidu Cloud
    • PaddlePaddle-based deployment
  • Hunyuan
    • Multimodal generation, including 3D assets
    • Integration with Tencent Cloud
  • Doubao
    • Multimodal agents via Volcengine Ark
    • High-throughput production workloads

Global Availability

Where can developers outside China actually reach these models, and how can they pay? This is one of the hardest things to find out — and one of the most useful.

Provider Global API Russia Middle East Southeast Asia Crypto Payment
DeepSeek ? ? ? ? —
Qwen ? ? ? ? —
GLM ? ◐ ◐ ? —
Kimi ◐ ? ? ◐ —
MiniMax ? ◐ ◐ ? —
ERNIE ◐ ? ? ◐ —
Hunyuan ◐ ? ? ◐ —
Doubao ◐ ? ◐ ? —

Legend: ✅ available · ◐ partial · — not available · ? unverified. Availability changes often. Always verify with the provider.

Benchmarks

This repo does not invent benchmarks. It points to established leaderboards and reports, with notes on how each covers Chinese AI models.

  • LMArena (Chatbot Arena) — Human-preference Elo rankings; includes major Chinese models.
    • Crowd-voted; useful for general quality signal, not task-specific performance.
  • Artificial Analysis — Quality, speed, and price comparisons across providers.
    • Good for cost and latency trade-offs; tracks Chinese models alongside global ones.
  • OpenCompass — Open-source evaluation suite with strong Chinese-language task coverage.
    • Maintained by Shanghai AI Lab; broad academic benchmark coverage.
  • SuperCLUE — Chinese-language general benchmark.
    • Chinese-language focus; verify test version and date.
  • Hugging Face Leaderboards — Task-specific community leaderboards (coding, reasoning, embeddings).
    • Coverage varies by leaderboard; check maintainer and methodology.

Latest Updates

See newsletters/latest.md for the full daily digest.

Model Providers

Featured Models

Entries from the machine-readable registry in data/models.yaml:

Model Provider Type Access Open Weights Official
DeepSeek-V3 DeepSeek chat, reasoning, coding api / web / weights / local-deploy / openai_compatible Yes Source
Qwen3 Alibaba Qwen chat, reasoning, coding api / web / weights / local-deploy / openai_compatible Yes Source
GLM-4 Zhipu AI chat, reasoning, coding api / web / openai_compatible No Source
Kimi Moonshot AI chat, long-context api / web / openai_compatible No Source
Qwen-Code Alibaba Qwen coding, agent-tool-use weights / local-deploy Yes Source
HY-World 2.0 Tencent Hunyuan image-generation, video-generation weights / local-deploy Yes Source
MiniCPM-V 4 OpenBMB (MiniCPM) chat, vision-language, reasoning weights / local-deploy Yes Source
Intern-S2-Preview InternLM (Shanghai AI Lab) chat, vision-language, reasoning weights / local-deploy Yes Source
Qwen3.6 Alibaba Qwen chat, reasoning, coding weights / local-deploy Yes Source
VoxCPM OpenBMB (MiniCPM) audio-speech weights / local-deploy Yes Source
Hy-MT2 Tencent Hunyuan translation weights / local-deploy Yes Source
HunyuanOCR Tencent Hunyuan ocr weights / local-deploy Yes Source

Collections

Machine-readable Data

The canonical registry lives in data/:

  • data/models.yaml — model index
  • data/providers.yaml — provider index
  • data/comparison.yaml — model comparison matrix
  • data/access_methods.yaml — how-to-access options
  • data/best_for.yaml — best-for mapping
  • data/availability.yaml — global availability matrix
  • data/benchmarks.yaml — aggregated benchmark sources
  • data/faq.yaml — frequently asked questions
  • data/news.yaml — daily updates and news events
  • data/resources.yaml — official resources, docs, model cards, tutorials
  • data/capabilities.yaml — normalized capability taxonomy
  • data/use_cases.yaml — reproducible playbooks and workflows

Run local validation:

python3 scripts/validate_data.py
python3 scripts/generate_readme.py --check

FAQ

What are Chinese AI models?

AI models developed by organizations based in China, such as DeepSeek, Qwen (Alibaba), GLM (Zhipu AI), Kimi (Moonshot AI), MiniMax, ERNIE (Baidu), and Hunyuan (Tencent). Many are released with open weights and expose OpenAI-compatible APIs.

Which Chinese AI model is best for coding?

DeepSeek and Qwen are commonly used for coding workloads, often as a lower-cost alternative to closed models. Compare them on a coding benchmark and on your own tasks before committing.

Are Chinese AI models open source?

Many are open-weight: DeepSeek, Qwen, GLM, and MiniMax publish weights on Hugging Face and ModelScope. Some, such as Kimi and Doubao, are API-only. Check the license of each specific model before deployment.

How can I access Chinese AI models from outside China?

Three common paths: call the provider's official API, use an OpenAI-compatible aggregator such as TheRouter or OpenRouter, or self-host open-weight models. See the "How to Access" section.

Are Chinese AI models OpenAI-compatible?

Most major providers expose an OpenAI-compatible API mode, so existing OpenAI SDK code usually works with a changed base URL and API key. Confirm supported parameters in each provider's documentation.

Are Chinese AI models cheaper than GPT or Claude?

Several Chinese models are priced below comparable closed models, which is why developers evaluate them as cost-efficient alternatives. Pricing changes frequently — verify current rates on the provider's pricing page.

Contributing

Please read CONTRIBUTING.md before opening an issue or PR.

Minimum requirements:

  • Prefer official sources: websites, docs, GitHub, Hugging Face, ModelScope, papers, or changelogs.
  • Include source URL, category, date, and why the entry belongs here.
  • Keep descriptions factual and neutral. Avoid unverified claims such as “最强”, “吊打”, or “全球第一”.
  • Do not submit random mirrors, SEO spam, unverifiable screenshots, cracked products, or gray-market workflows.

TheRouter

TheRouter.ai provides an OpenAI-compatible access layer for multiple AI models, with routing, fallback, and cost-control workflows. It is listed in How to Access as one option among several. This repo stays source-driven and vendor-neutral; entries must remain community-useful.

Disclaimer

This is a community-maintained index. Always verify model availability, pricing, licensing, and benchmark claims with the official provider.

About

Curated index of AI models developed in China, plus discovery notes and links.

therouter.ai

Topics

awesome-list ai-models model-catalog china-models

Resources

Readme

Contributing

Contributing
Activity
Custom properties

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages