Slow responses on mainland China networks: testing RelayRouter direct connectivity

To test whether RelayRouter improves slow responses on a mainland China network, point your existing SDK at the RelayRouter base URL, send identical requests to the same model through your current route and through RelayRouter, and compare start transfer time and total time. RelayRouter's published facts do not include latency benchmarks, so measurements taken on your own network are the evidence to rely on when deciding whether to switch.

Why a side by side test needs no code changes

You can run the test by changing the base URL and the API key in your existing client. According to the relayrouter.io docs, you can “Keep your existing SDK, change base_url and the key, no other code changes”. This keeps the comparison controlled: the request body, model name and client code stay identical, so differences in response time come from the network path and upstream handling rather than from your code. For OpenAI SDKs the base URL is https://relayrouter.io/v1, and for Anthropic SDKs it is https://relayrouter.io. Authentication uses the header Authorization: Bearer YOUR_API_KEY, and keys are created in the dashboard at https://relayrouter.io/dashboard. Payment is by Stripe card. Full setup details are in the RelayRouter docs.

How to run a direct connectivity test

A practical test measures connection time, start transfer time and total time for the same request over both routes.

1. Create an API key at https://relayrouter.io/dashboard.
2. Choose one model ID, for example gemini-3.8-flash or gpt-5.6-sol, and a short fixed prompt.
3. Send a timed request from a machine on your mainland China network:

   curl -s -o /dev/null -w "connect: %{time_connect}s start: %{time_starttransfer}s total: %{time_total}s\n" https://relayrouter.io/v1/chat/completions -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"model":"gemini-3.8-flash","messages":[{"role":"user","content":"Reply with OK"}]}'

4. Repeat the request several times at different hours and record the median values.
5. Run the same script against your current endpoint and compare the medians. Enable streaming to judge how quickly output begins to arrive.

Which protocols and models you can test

RelayRouter lets you test about 108 models across 19 public groups through one gateway. According to relayrouter.io, the service is “Compatible with the OpenAI, Anthropic and Gemini protocols”, using POST /v1/chat/completions, POST /v1/messages and POST /v1beta/models/{model}:generateContent respectively. This means you can test the client type your application already uses instead of rewriting requests for a different format. Covered models include the Claude family (including claude-opus-5-5, claude-fable-5-1 and claude-opus-5), GPT-6 and GPT-5.6 (gpt-6-astra, gpt-5.6-sol), Gemini 3.8 Flash (gemini-3.8-flash), plus DeepSeek, GLM, MiniMax and Moonshot. Streaming is supported, which helps when perceived responsiveness matters more than total completion time. The current list is on the RelayRouter models page.

What a connectivity test costs

Running tests through RelayRouter costs the model usage itself, because there is a $0 platform fee, no minimum spend and no subscription. Failed or errored requests are generally not billed, which is relevant if some test requests time out on a slow network. 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. For direct pricing, deepseek-v4-flash is billed at 1.1x DeepSeek official time-of-day prices: off-peak rates are CNY 1.1 per 1M input tokens and CNY 4.4 per 1M output tokens, and these double on weekdays from 09:00 to 12:00 and 14:00 to 18:00 Beijing time. Scheduling DeepSeek tests outside those windows lowers their cost. Live rates are on the models page.

FAQ

Do I need a new SDK to test RelayRouter from mainland China?

No. Keep your existing OpenAI, Anthropic or Gemini SDK and change the base URL and API key.

Does RelayRouter publish latency numbers for mainland China?

The published facts do not include latency figures, so run the timed curl test above on your own network and compare median values against your current route.

Will timed out test requests be charged?

Failed or errored requests are generally not billed, and there is a $0 platform fee, so test costs come from successful model usage at the rates listed on https://relayrouter.io/models.