Claude API Relay
Claude API Relay, Done Right
One account, one endpoint (/v1), Claude models alongside many more models — switch the base URL, keep your SDK.
1
Create an API key
Sign up and create a key in the console, with an optional budget.
2
Point at /v1
Set your client's base_url to https://apitokenrouter.com/v1.
3
Swap the config
Drop in the new key. Your existing SDK calls stay as they are.
Endpoint (OpenAI-compatible)
https://apitokenrouter.com/v1What a relay actually saves you
Relaying through one endpoint is not about saving a line of code. It is about what stops being your job:
- Smart routing — requests are steered across channels by state, not pinned to a single upstream;
- Automatic fallback — when a channel misbehaves the platform switches, and owns the retry logic;
- Prompt caching — repeated prefixes hit the cache instead of being billed twice;
- One bill — every model settles in one place, no cross-vendor reconciliation;
- Consumption tracking — usage per model is visible in the console.
Protocol compatibility
The endpoint follows the OpenAI protocol. Point your existing client at /v1 with your key and your current SDK calls keep working — same request shape, same response shape, same model naming you already use.
Billing and pricing
One unified bill with consumption tracking broken down by model. Per-model prices live on the model price list; prices at checkout and settlement prevail.
This page quotes no prices on purpose. A stale number on a marketing page is worse than no number — so pricing has exactly one verifiable source.
Claude clients we document
Step-by-step guides for the clients people actually run Claude through.
Claude Code
Terminal-first agent. Set ANTHROPIC_BASE_URL and your key, then keep using the same workflow.
Install guideFAQ
Is it compatible with the OpenAI SDK?
Yes. The endpoint follows the OpenAI protocol; point your existing client at /v1 with your API key.
Which Claude clients are supported?
Claude Code and other CLI/desktop clients; step-by-step guides live in our docs (many clients supported).
How is usage billed?
One unified bill with consumption tracking in the console; per-model prices are listed on the model price list page (/models.html); prices at checkout/settlement prevail.
What happens if a channel fails?
The platform performs smart scheduling with automatic fallback across channels.
Where do I see pricing?
On the model price list page (/models.html); this page does not quote prices.
Get a key and make one call
Creating an account and running your first request takes minutes, not a migration project.