How to Switch from OpenAI to Chinese LLMs in 5 Minutes

A complete migration guide — change 2 lines of code, save 80-99%

OpenAI's GPT-4o is excellent, but it's expensive. For many use cases, Chinese LLMs like DeepSeek V3 and Qwen3 deliver comparable or better quality at a fraction of the cost. This guide shows you exactly how to migrate in 5 minutes.

💰 Typical savings after migration:
• GPT-4o → DeepSeek-V3: Save 90% (same quality tier)
• GPT-4o-mini → Qwen3-Flash: Save 93% (better Chinese support)
• GPT-4o → GLM-4-Flash: Save 100% (it's free!)

Step 1: Get Your API Key

Sign up at AI Token Hub

Go to /get-started and get your free API key with 50K tokens. No credit card required. You'll receive an OpenAI-compatible API key instantly.

Step 2: Change Your Base URL

Update 2 Lines of Code

If you're using the OpenAI Python SDK, you only need to change the api_key and base_url:

from openai import OpenAI

# BEFORE (OpenAI)
# client = OpenAI(api_key="sk-proj-...")

# AFTER (AI Token Hub)
client = OpenAI(
    api_key="your-aitoken-key",
    base_url="https://eaf9553505eeb8f5-115-190-107-107.serveousercontent.com/v1"
)

That's it! All your existing code works unchanged.

Step 3: Choose Your Model

Map OpenAI Models to Chinese Equivalents

OpenAI Model	→ Replace With	Savings
GPT-4o	deepseek-v3	90%
GPT-4o-mini	qwen3-flash	93%
GPT-4o (reasoning)	deepseek-r1	85%
GPT-3.5-turbo	glm-4-flash	100% (FREE)

response = client.chat.completions.create(
    model="deepseek-v3",  # Was "gpt-4o"
    messages=[{"role": "user", "content": "Hello!"}],
    temperature=0.7
)
# Same API, same parameters, 90% cheaper!

Step 4: Test & Validate

Run Your Test Suite

Run your existing tests with the new model. Most applications see identical or improved results. Key areas to verify:

Response format and structure (identical with OpenAI-compatible API)
Content quality (DeepSeek V3 matches GPT-4o on most benchmarks)
Latency (Chinese models are often faster due to less global demand)
Token counting (same tiktoken library works)

Step 5: Optimize with Model Routing

Use Different Models for Different Tasks

# Smart routing: use cheap models for simple tasks
def smart_call(prompt, complexity="medium"):
    if complexity == "low":
        model = "glm-4-flash"       # FREE
    elif complexity == "medium":
        model = "qwen3-flash"       # $0.01/M
    else:
        model = "deepseek-v3"       # $0.27/M
    
    return client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": prompt}]
    )

LangChain Migration

If You Use LangChain

from langchain_openai import ChatOpenAI

# Just change these parameters
llm = ChatOpenAI(
    model="deepseek-v3",
    api_key="your-aitoken-key",
    base_url="https://eaf9553505eeb8f5-115-190-107-107.serveousercontent.com/v1"
)

# Everything else works the same
result = llm.invoke("What is 2+2?")

Common Concerns

"Are Chinese LLMs as good as GPT-4o?"
DeepSeek-V3 matches GPT-4o on MMLU, HumanEval, and GSM8K benchmarks. For Chinese language tasks, it often outperforms GPT-4o. For English-only tasks, the quality difference is minimal.

"What about data privacy?"
AI Token Hub routes through international endpoints. No data is stored or logged. API keys are scoped per account.

"Can I switch back?"
Yes! Since we use the OpenAI-compatible API, you can switch between providers by changing the base_url. Keep both configurations and switch as needed.

Start Free Migration — 50K Tokens →