DEV Community

ryanzhaowowryan-ui
ryanzhaowowryan-ui

Posted on

How to Use DeepSeek, Qwen and Seedance APIs from Outside China (OpenAI-Compatible)

#python #ai #webdev #api

Chinese frontier models like DeepSeek, Qwen, GLM and the Seedance 2.0 video model have become genuinely competitive on price and quality. The problem for most of us isn't the models — it's access. Sign-up often expects a mainland China phone number, billing wants a domestic payment method, and some endpoints rate-limit or block traffic from outside the country.

This post shows a clean way to call these models from anywhere using an OpenAI-compatible gateway, so you keep your existing SDK and only change three things: the base URL, the API key, and the model name.

The OpenAI-compatible trick

If your code already talks to OpenAI, you're 90% done. An OpenAI-compatible endpoint accepts the same /v1/chat/completions request shape. So you point your client at a different base URL and pick a Chinese model.

I've been using ComputeBridge for this — it exposes DeepSeek, Qwen, GLM and Seedance behind one OpenAI-compatible endpoint, bills in USD, and doesn't need a China phone number. (There are other gateways; the pattern below works for any OpenAI-compatible provider.)

Python (openai SDK) 

from openai import OpenAI

client = OpenAI(
    base_url="https://www.computebridge.top/v1",
    api_key="YOUR_API_KEY",
)

resp = client.chat.completions.create(
    model="deepseek-v4-pro",          # or qwen3.7-max, glm-5.1, ...
    messages=[{"role": "user", "content": "Explain mixture-of-experts in 2 sentences."}],
)
print(resp.choices[0].message.content)
Enter fullscreen mode Exit fullscreen mode

Node (openai SDK) 

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://www.computebridge.top/v1",
  apiKey: process.env.COMPUTEBRIDGE_API_KEY,
});

const r = await client.chat.completions.create({
  model: "qwen3.7-max",
  messages: [{ role: "user", content: "Translate 'good morning' into Bahasa Indonesia." }],
});
console.log(r.choices[0].message.content);
Enter fullscreen mode Exit fullscreen mode

curl 

curl https://www.computebridge.top/v1/chat/completions \
  -H "Authorization: Bearer $COMPUTEBRIDGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Say hi."}]
  }'
Enter fullscreen mode Exit fullscreen mode

Which model for what

  • DeepSeek (deepseek-v4-pro, deepseek-v4-flash) — reasoning, coding, structured output, agents. Usually the cheapest strong option.
  • Qwen (qwen3.7-max, qwen3-coder-plus, …) — multilingual generation and coding; great for Southeast Asian languages.
  • GLM (glm-5.1) — agents, coding, general chat; useful for model diversity.
  • Seedance 2.0 (doubao-seedance-2-0, …-fast) — text-to-video generation, via the task API.

Tips

  • Use a separate API key per app so you can rotate and track usage.
  • Keep model names explicit — don't rely on a gateway silently substituting a high-end name for a cheaper one.
  • Benchmark with your prompts. Generic leaderboards rarely match your workload; cost-per-task is what matters.

Wrapping up

If you've been putting off trying Chinese models because of the access friction, the OpenAI-compatible route removes it entirely. Swap the base URL, drop in a key, and you can A/B DeepSeek against GPT or Claude in an afternoon.

Full model list and pricing: https://www.computebridge.top/pricing
Quickstart docs: https://www.computebridge.top/docs/quickstart

What Chinese models have you tried, and how did they compare on your tasks? Curious to hear in the comments.

Top comments (0)