LiteLLM integration

Use LiteLLM with Chinese AI models through ChinaAPI.

Add ChinaAPI as an OpenAI-compatible backend in LiteLLM proxy or the Python SDK, then route applications to Chinese models with one key.

Start free — $2 credit Live model pricing

Prerequisites

Before editing config.

LiteLLM installed

Use either the LiteLLM proxy server or the Python SDK.

ChinaAPI endpoint

Base URL is https://api.chinaapi.ai/v1. API keys are created under Dashboard -> Console -> Tokens.

Proxy setup

Use `openai/<model>` entries in `config.yaml`.

model_list:
  - model_name: deepseek-fast
    litellm_params:
      model: openai/deepseek-flash
      api_base: https://api.chinaapi.ai/v1
      api_key: os.environ/CHINAAPI_API_KEY

  - model_name: kimi-code
    litellm_params:
      model: openai/kimi-k2.7-code
      api_base: https://api.chinaapi.ai/v1
      api_key: os.environ/CHINAAPI_API_KEY

  - model_name: qwen-max
    litellm_params:
      model: openai/qwen3.7-max
      api_base: https://api.chinaapi.ai/v1
      api_key: os.environ/CHINAAPI_API_KEY

export CHINAAPI_API_KEY="sk-..."
litellm --config config.yaml --port 4000

Python SDK

Call ChinaAPI directly from LiteLLM.

import os
from litellm import completion

response = completion(
    model="openai/deepseek-flash",
    api_base="https://api.chinaapi.ai/v1",
    api_key=os.environ["CHINAAPI_API_KEY"],
    messages=[{"role": "user", "content": "Say LiteLLM is connected to ChinaAPI."}],
    stream=True,
)

for chunk in response:
    print(chunk, end="")

Recommended models

Good LiteLLM routes.

For current model details and rate evidence, review DeepSeek-V4 Flash and Kimi K2.7 Code before setting a production route.

`deepseek-flash`

Fast fallback route for chat, coding help, and agent loops.

`kimi-k2.7-code`

Good coding route for repository-level tasks and longer sessions.

`qwen3.7-max`

Reasoning-heavy route when quality matters more than latency.

`glm-5.2`

Long-context route for larger documents and codebases.

`MiniMax-M3`

Multimodal coding and long-context model for alternate routing.

Troubleshooting

Common LiteLLM issues.

Thinking models such as qwen3.7-max can generate reasoning tokens before final output. Use streaming in clients behind LiteLLM or raise upstream/client timeouts.

Model not found: check the exact model ID on live pricing, then update litellm_params.model. Keep the openai/ prefix for OpenAI-compatible routing.

401: confirm CHINAAPI_API_KEY resolves to an sk-... key and LiteLLM is passing it as the provider API key.

Pricing and signup

Use gateway-synchronized token pricing behind your LiteLLM gateway.

Standard model rates follow providers' international list prices at 100%; some models have a separate Paid rate after the first top-up. Register for $2 free credit, then check live pricing before routing production traffic.