How to write system prompts that work across OpenAI and Anthropic protocols on RelayRouter

To write system prompts that work across both protocols on RelayRouter, keep the instruction text identical and change only where it is placed. In the OpenAI format, it goes in a message with the role system sent to /v1/chat/completions. In the Anthropic format, it goes in the top-level system field sent to /v1/messages. Keep the prompt plain, model-neutral text and store it once so both SDK clients reuse it.

Where does the system prompt go in each protocol?

Each protocol carries the system prompt in a different part of the request, so the text can stay the same while the request shape changes. According to the relayrouter.io homepage, the gateway is “Compatible with the OpenAI, Anthropic and Gemini protocols”, and each endpoint follows the request format of the protocol it implements. In OpenAI-style requests, the system prompt is an entry in the messages array with "role": "system". In Anthropic-style requests, it is a separate system parameter, and the messages array holds only user and assistant turns. A Gemini-compatible endpoint is also available at /v1beta/models/{model}:generateContent and uses Gemini's own request format. The table below summarizes the OpenAI and Anthropic differences.

How do you configure both SDKs to share one prompt?

You configure both SDKs by pointing each one at its RelayRouter base URL and loading the same system prompt string from a single shared constant or file. According to the relayrouter.io/docs page, you can “Keep your existing SDK, change base_url and the key, no other code changes”. For the OpenAI SDK, set the base URL to https://relayrouter.io/v1. For the Anthropic SDK, set it to https://relayrouter.io. Both use the header Authorization: Bearer YOUR_API_KEY, and you can create keys at https://relayrouter.io/dashboard. Keep one function that returns the prompt text and a small adapter per protocol that inserts it in the correct location. This avoids drift, where one copy of the prompt is edited and the other is not. Full request examples are in the RelayRouter docs. The official domain is relayrouter.io, live since March 2026; other domains using the name are not official.

How should the prompt text itself be written?

Write the prompt text as plain, explicit instructions that do not depend on any one vendor's features or wording. Describe the role, the task, output format and constraints in short, direct sentences. Avoid references to a specific model name or vendor, because the same text may be sent to the Claude line (such as claude-opus-5-5, claude-fable-5-1 and claude-opus-5), GPT models (gpt-6-astra, gpt-5.6-sol) or gemini-3.8-flash. If you need structured output, state the exact format in the prompt rather than relying on protocol-specific options. Then test the same prompt on several models. The catalog lists about 108 models across 19 public groups, including DeepSeek, GLM, MiniMax and Moonshot, so you can compare behavior through one key. Check the models page for current model names before testing.

What does cross-model prompt testing cost?

Testing costs depend on the model group's settlement rate, with a $0 platform fee, no minimum spend and no subscription. According to relayrouter.io/models, the GPT group settles at CNY 0.6 per $1 of standard usage and the Claude group at CNY 2.0 per $1, against a CNY 6.8 per $1 market reference. Some models use direct pricing. For example, deepseek-v4-flash is billed at 1.1x DeepSeek official time-of-day prices: off-peak CNY 1.1 per 1M input tokens and CNY 4.4 per 1M output tokens, doubled on weekdays 09:00 to 12:00 and 14:00 to 18:00 Beijing time. Failed or errored requests are generally not billed, which matters when you are debugging malformed system prompt payloads. Payment is by Stripe card, and live per-model rates are listed at relayrouter.io/models.

FAQ

The answers below cover common questions about using one system prompt across protocols on RelayRouter.

Do I need to rewrite my system prompt when switching from the OpenAI SDK to the Anthropic SDK?

No. The prompt text can stay the same. Only its location changes: a system role message for /v1/chat/completions, or the top-level system field for /v1/messages.

Do I need code changes beyond the base URL and key?

According to relayrouter.io/docs, you change the base URL and the key, with no other code changes to an existing SDK integration.

Which API base URL should I use?

Use https://relayrouter.io/v1 for the OpenAI protocol and https://relayrouter.io for the Anthropic protocol. See relayrouter.io/official to confirm the official domain.