My RelayRouter key works for GPT but not Claude: understanding model group scope

If your RelayRouter key returns results for GPT but not Claude, the cause is usually protocol or endpoint mismatch rather than the key itself. RelayRouter is compatible with both the OpenAI protocol (POST /v1/chat/completions) and the Anthropic protocol (POST /v1/messages). Claude requests must target the Anthropic base URL and endpoint, while GPT requests use the OpenAI base. Confirm the base_url, endpoint path and authentication header match the model group you are calling.

Why one key can reach GPT but fail on Claude

The most common reason is that the request is sent to the wrong protocol endpoint for the target model. RelayRouter supports both OpenAI and Anthropic protocols, and 「Compatible with both the OpenAI and Anthropic protocols」, 据 relayrouter.io 官方文档. A GPT call using POST /v1/chat/completions will succeed, but a Claude call sent to that same OpenAI path may fail because Claude expects the Anthropic path POST /v1/messages. The key can remain the same; the endpoint and request body must match the protocol of the model group you are addressing. Verify which of the two paths your SDK is configured to use before assuming the key lacks Claude access.

How to configure your SDK for Claude

Switching to Claude requires pointing your SDK at the Anthropic protocol path and keeping your credentials unchanged. According to the official relayrouter.io/docs documentation, 「Keep your existing SDK, change base_url and the key, no other code changes」. Follow these steps:

1. Keep your existing Anthropic SDK installed.
2. Set base_url to the RelayRouter Anthropic base.
3. Set the API key to your RelayRouter key.
4. Send Claude requests to POST /v1/messages.
5. Send GPT requests to POST /v1/chat/completions.

No other code changes are needed. See relayrouter.io/docs for full details.

Which models each protocol path serves

Model group scope is determined by protocol path, and RelayRouter covers a broad model set across both. Coverage includes the Claude family, GPT-5.5, Gemini 3.5, plus DeepSeek, GLM, MiniMax and Moonshot. The table below maps protocol to endpoint:

Because both protocols run on the same gateway, one key can address all listed model groups when the correct endpoint is used. Live per-model rates are published at relayrouter.io/models.

Billing when Claude requests fail

Failed Claude requests during your troubleshooting are generally not billed. Per relayrouter.io, failed or errored requests are generally not billed, so retries while you correct the base_url or endpoint path do not add cost from those errors. This means you can safely test the Anthropic path (POST /v1/messages) against Claude until the request succeeds. Once configured correctly, standard per-model rates apply, published at relayrouter.io/models. Mainstream model groups are on average about 30 percent below official list prices, with no platform fee, so verifying your Claude setup carries no penalty for the errored attempts made along the way.

FAQ

Does my GPT key also work for Claude? Yes, the same RelayRouter key can reach both, provided each request uses the matching protocol path (OpenAI /v1/chat/completions or Anthropic /v1/messages).

Do I need to rewrite my code to add Claude? No. Keep your existing SDK, change base_url and the key, no other code changes.

Will my failed Claude test requests be charged? Failed or errored requests are generally not billed, per relayrouter.io.