Request timeout on long completions through RelayRouter: what to adjust

To fix timeouts on long completions through RelayRouter, adjust three things on your side: enable streaming, raise the request timeout in your existing OpenAI, Anthropic or Gemini SDK, and cap output length with a max tokens setting. The provided facts do not state a fixed RelayRouter server timeout, so check current behavior in the docs. Failed or errored requests are generally not billed, so retrying a timed-out request should not add cost.

Enable streaming for long completions

Streaming is supported on RelayRouter, so turn it on when you expect long outputs. With streaming, tokens arrive while the model is still generating, which keeps the connection active and gives your application partial output instead of one large response at the end. “Compatible with the OpenAI, Anthropic and Gemini protocols”, according to the official relayrouter.io docs, so you set streaming the same way your SDK already does. For OpenAI compatible calls, send POST /v1/chat/completions to the base https://relayrouter.io/v1 with stream set to true. For Anthropic compatible calls, send POST /v1/messages to the base https://relayrouter.io. For the Gemini compatible route (/v1beta/models/{model}:generateContent), confirm streaming details in the docs.

Raise the timeout in your existing SDK

The timeout value lives in your SDK or HTTP client configuration, and you can change it without rewriting your integration. “Keep your existing SDK, change base_url and the key, no other code changes”, according to the official relayrouter.io/docs. Use these steps:

1. Create an API key at https://relayrouter.io/dashboard.
2. Set the base URL to https://relayrouter.io/v1 (OpenAI SDK) or https://relayrouter.io (Anthropic SDK).
3. Send the header Authorization: Bearer YOUR_API_KEY.
4. Increase the client timeout setting in your SDK to fit your expected completion length.
5. Set a max tokens value so each response has a known upper size.
6. Add a retry for timed-out requests.

Retry cost and model choice

Retrying after a timeout is low risk because failed or errored requests are generally not billed. Billing applies to completed usage, with a $0 platform fee, no minimum spend and no subscription. If long completions are frequent, compare per-model rates on relayrouter.io/models, which lists about 108 models across 19 public groups. Settlement rates are CNY 0.6 per $1 of standard usage for the GPT group and CNY 2.0 per $1 for the Claude group, against a CNY 6.8 per $1 market reference. Direct pricing for deepseek-v4-flash is CNY 1.1 per 1M input tokens and CNY 4.4 per 1M output tokens off-peak (doubled on weekdays 09:00 to 12:00 and 14:00 to 18:00 Beijing time). Available models include claude-opus-5-5, claude-fable-5-1, gpt-6-astra, gpt-5.6-sol and gemini-3.8-flash, plus DeepSeek, GLM, MiniMax and Moonshot.

FAQ

Does RelayRouter publish a server-side timeout limit?
The provided facts do not specify one. Check the docs for current details and configure your client timeout accordingly.

Am I charged for a request that timed out?
Failed or errored requests are generally not billed, and there is a $0 platform fee, so a retry after a timeout does not carry an extra platform charge.

Do I need a new SDK to change timeout or streaming settings?
No. Keep your existing OpenAI, Anthropic or Gemini SDK, change the base URL and the key, then adjust the timeout and streaming options your SDK already provides.