One key,
all major AI models.
Claude ·
GPT ·
DeepSeek ·
Doubao ·
Qwen.
Swap one base_url and you are in. Compatible with both Anthropic and OpenAI SDKs. Pay-as-you-go.
// Get an API key on signup. Pay-as-you-go, no upfront fee.
Why Relay
Use the world's top AI models from China as if they were native APIs.
China-direct
Servers run on routes reachable from China. No VPN, no proxy tools, low latency, stable connection.
Dual protocol
Supports both Anthropic SDK (/v1/messages) and OpenAI SDK (/v1/chat/completions). ChatBox, LobeChat, Cursor — just change the base URL.
Multi-vendor models
Covers Claude Sonnet 4.5 / Opus 4.7 / Haiku 4.5, GPT-5.5 / 5.4 series, DeepSeek V4 Flash/Pro, Doubao, Qwen full series. Routed by model name.
Reliable
Multi-key load balancing on the upstream side. Auto-failover when a single key has issues.
One line to integrate
Use whichever SDK fits — Relay routes to the right upstream automatically. One key, one base_url change, the rest of your code stays untouched.
- Dual protocol: Anthropic /v1/messages + OpenAI /v1/chat/completions
- Native support for major SDKs: anthropic / openai / langchain / autogen
- Works with ChatBox, LobeChat, Cherry Studio, Cursor and more
- Streaming (SSE) fully supported, error format matches the official one
// CLIENTS
Using a client app?
In your client settings, change the API endpoint to Relay and paste your Relay key. No coding required.
# Anthropic SDK — call Claude models
import anthropic
client = anthropic.Anthropic(
base_url="https://your-relay.com",
api_key="rk-your-relay-key",
)
message = client.messages.create(
model="claude-sonnet-4-5-20250929",
max_tokens=1024,
messages=[{"role": "user", "content": "hello!"}]
)# OpenAI SDK — call GPT / DeepSeek / Doubao
from openai import OpenAI
client = OpenAI(
base_url="https://your-relay.com/v1",
api_key="rk-your-relay-key",
)
resp = client.chat.completions.create(
model="deepseek-v4-flash", # or gpt-5.4, doubao-seed-2-0-pro-260215
messages=[{"role": "user", "content": "hello!"}]
)Transparent usage-based pricing
Pay for what you use. No monthly fee, no minimum spend.
| Model | Provider | Input / MTok | Output / MTok |
|---|---|---|---|
| DeepSeek V4 Flash Best value | DeepSeek | $0.17 | $0.34 |
| Qwen Turbo CN model | Qwen | $0.05 | $0.11 |
| Claude Sonnet 4.5 Recommended | Anthropic | $3.60 | $18.00 |
| GPT-5.4 General | OpenAI | $3.00 | $18.00 |
| Claude Opus 4.7 Flagship | Anthropic | $6.00 | $30.00 |
// Sample pricing = vendor list price × 1.3. Actual markup is configured by the operator and applied at billing time. MTok = million tokens.
// Sample × 1.3. MTok = million tokens. Actual price applies at billing time.
- ✓ Sign up and start, no prepayment
- ✓ One key works across multiple model vendors
- ✓ Real-time usage view, multi-dimensional analytics