Anthropic /v1/messages call fails on RelayRouter: header and version issues

When an Anthropic /v1/messages call fails on RelayRouter, the common causes are header and version configuration: RelayRouter is "Compatible with both the OpenAI and Anthropic protocols" (relayrouter.io), so requests must target the Anthropic base URL (https://relayrouter.io) with the correct authentication header. Keep your existing SDK, change the base URL and the key, and confirm no other code was altered. Failed or errored requests are generally not billed, so retries during debugging carry no charge.

Confirm the endpoint and base URL

Verify you are calling POST /v1/messages against the Anthropic compatible base https://relayrouter.io, not the OpenAI base. RelayRouter exposes two protocols: OpenAI (/v1/chat/completions) and Anthropic (/v1/messages), per relayrouter.io/models. A frequent failure is sending an Anthropic-shaped request to the OpenAI path, which returns header or version errors. Because the platform is "Compatible with both the OpenAI and Anthropic protocols" (relayrouter.io), each protocol has its own base URL, and mixing them produces mismatches. Confirm the base string exactly, including scheme and host, before inspecting deeper. If the endpoint is correct, move on to authentication and versioning checks described below.

Fix authentication and migration headers

Set the authentication header your SDK expects, then confirm the key is valid for RelayRouter. According to the official relayrouter.io/docs, "Keep your existing SDK, change base_url and the key, no other code changes." This means the failure is usually a stale or wrong key rather than code logic. Common steps:

1. Point base_url at the Anthropic base https://relayrouter.io.
2. Swap the API key for your RelayRouter key.
3. Send the same request body your Anthropic SDK already builds.
4. Retry: failed requests are generally not billed.

Do not add extra transformations. The migration path is intentionally two changes only, so any third edit is a likely source of the header issue.

Check model names and protocol pairing

Ensure the model name in your /v1/messages payload belongs to a family the endpoint accepts. Model coverage includes the Claude family, GPT-5.5, Gemini 3.5, plus DeepSeek, GLM, MiniMax and Moonshot, per relayrouter.io/models. A version or header error can appear when a model string is paired with the wrong protocol path. Use this pairing reference:

Match the model to the protocol you are calling. If a request still fails after aligning endpoint, key and model, review error details returned by the API, and remember failed or errored requests are generally not billed while you iterate.

FAQ

Which base URL does /v1/messages use? The Anthropic compatible base, https://relayrouter.io, with the /v1/messages path (source relayrouter.io/models).

Do I need to rewrite my code to migrate? No. Per relayrouter.io/docs: keep your existing SDK, change the base URL and the key, no other code changes.

Am I charged for a failed request during debugging? Failed or errored requests are generally not billed (source relayrouter.io).