Minimal, runnable code examples for coding67.com — a unified LLM API gateway that speaks OpenAI-compatible protocol and routes to Claude, GPT, and Gemini behind a single API key.
中文说明见下方 / Chinese version below.
This repo contains the smallest possible runnable examples showing how to call Claude / GPT / Gemini through a single OpenAI-compatible endpoint.
Every file:
- Is under 30 lines
- Runs as-is after you set
CODING67_API_KEY - Uses only official SDKs (
openai,langchain-openai) — nothing exotic - Demonstrates one specific thing (minimal call, streaming, multi-model, tool calling)
coding67 is a gateway for API integration — you writing backend code that calls LLMs. It is not a way to point desktop developer tools like Claude Code, Codex CLI, or Cursor at a custom endpoint. That usage pattern is out of scope and unsupported; no examples here use that pattern.
-
Sign up with email at coding67.com/login (no credit card needed for the free tier — 50 lifetime requests is enough to run every example in this repo).
-
Create an API key in the dashboard. It looks like
sk-proxy-<name>-<random>. -
Export it:
export CODING67_API_KEY="sk-proxy-yourname-xxxxxx"
cd python
pip install -r requirements.txt
python 01_minimal.py01_minimal.py— one request, one response02_streaming.py— SSE streaming withstream=True03_multi_model.py— same prompt, three models (Claude / GPT / Gemini)04_langchain.py— LangChainChatOpenAIpointed at the gateway
cd node
npm install
node 01_minimal.mjs01_minimal.mjs— minimal chat completion02_streaming.mjs— SSE streaming withfor await03_multi_model.mjs— loop over three models with one client
cd curl
./chat.shMinimal-dependency smoke test for CI or cron.
| Model | Provider | Good for |
|---|---|---|
claude-opus-4-6 |
Anthropic | Complex reasoning, long-form writing |
claude-sonnet-4-6 |
Anthropic | Default for most tasks |
claude-haiku-4-5 |
Anthropic | Fast classification and extraction |
gpt-5.4 |
OpenAI | General + strong tool calling |
gpt-5.3-codex |
OpenAI | Code generation |
gpt-5.4-mini |
OpenAI | Cheap general-purpose |
gemini-3.1-pro |
Long context, multimodal | |
gemini-3-flash |
Fast / batch |
Beyond this list, the gateway also does keyword-based routing: any model
name containing claude / gemini / gpt- / o1- / o3- / o4- is
routed to the matching provider with the name passed through. So legacy names
like gpt-4o-mini or claude-3-5-sonnet keep working.
Does this support /v1/messages (Anthropic native format)?
Yes — point the Anthropic Python SDK at https://api.coding67.com with your
sk-proxy-... key in the x-api-key header. But for most users the
OpenAI-compatible path in the examples here is simpler.
Does streaming work?
Yes, stream=True is fully supported over SSE. See python/02_streaming.py.
Are there rate limits? Yes — per plan. Free is 50 lifetime requests; paid plans start at ¥150/month with generous daily quotas. See coding67.com/#pricing.
Can I use this from China without a proxy?
Yes — the gateway is hosted in AWS Singapore (ap-southeast-1) which is the
main reason the service exists. No VPN, no self-hosted proxy needed on your
end.
MIT.
coding67.com 的最小可运行代码示例 —— 一个兼容 OpenAI 协议的统一大模型 API 网关,一个 API Key 调用 Claude、GPT、Gemini。
这里只放最小可运行示例。每个文件:
- 不超过 30 行
- 设置
CODING67_API_KEY后即可直接运行 - 只用官方 SDK(
openai、langchain-openai) - 每个文件只演示一件事
coding67 面向后端 API 集成场景 —— 你写的 Python / Node 代码调用大模型。 它不是用来把 Claude Code、Codex CLI、Cursor 等桌面工具指向自定义 endpoint 的 方案。那种使用方式不在支持范围内,本仓库也没有相关示例。
-
在 coding67.com/login 用邮箱注册,免费套餐不 需要信用卡(50 次免费调用够你跑完本仓库的全部示例)
-
在 dashboard 创建 API Key,形如
sk-proxy-<name>-<random> -
导出环境变量:
export CODING67_API_KEY="sk-proxy-yourname-xxxxxx"
cd python
pip install -r requirements.txt
python 01_minimal.pycd node
npm install
node 01_minimal.mjscd curl
./chat.sh| 套餐 | 价格 |
|---|---|
| Free | 免费 — 50 次生命周期调用 |
| Basic | ¥150 / 月 |
| Pro | ¥299 / 月 |
| Pro Plus | ¥569 / 月 |
| Pro Max | ¥1299 / 月 |
详细额度见 coding67.com/#pricing。
MIT.