China latency benchmark
Router One median latency: ~120ms p50 from Beijing / Shanghai / Shenzhen. VPN-direct to upstream LLM providers: ~620ms p50 with 8–15% timeout rate.
These numbers are what drive the latency claims used elsewhere on the site. Each figure carries its measurement date so the page is verifiable, not aspirational. Method details live on the methodology page.
Last updated: · Update cadence: re-measured when routes or the model catalog change
| City | VPN-direct p50 (timeout %) | Router One p50 (timeout %) |
|---|---|---|
| Shanghai | 620ms (12%) | 120ms (0%) |
| Beijing | 580ms (8%) | 110ms (0%) |
| Shenzhen | 700ms (15%) | 130ms (0%) |
What we measure
- Probe shape
- End-to-end HTTPS request to api.router.one with a representative chat completion payload (~500 input tokens, ~200 output tokens, non-streaming).
- Window
- Rolling 7 days at measurement time. The measurement date is shown at the top of this page and re-runs are published here first.
- Publication scope
- This table is a published snapshot for the rolling 7-day window ending on the last updated date. It is not an SLA; contractual availability is documented on /sla.
- Sample size
- ≥ 5,000 requests per city × ISP combination per week. Outliers above p99.5 are reported separately and not folded into the headline median.
- Networks
- Each city runs probes on China Telecom, China Unicom, and China Mobile.
Caveats and reading the table
- VPN comparison baseline
- The 'VPN-direct' row represents calling upstream LLM provider endpoints from China through a typical commercial VPN. Your specific VPN and ISP combination may differ.
- Median, not best case
- Numbers are p50. Tail behaviour (p95, p99) is published in the dashboard for active accounts.
- Region-specific values
- The headline on /claude-code-china and other landing pages uses the median across the three cities and three ISPs. Region-specific pages should label their numbers explicitly.
FAQ
- Can I reproduce these numbers?
- Yes — probe shape, sample size, and ISP set are documented on /methodology. Each published figure on /benchmarks/china-latency is tied to the measurement date shown at the top of the page.
- Why is VPN latency this high?
- Cross-border TCP/TLS handshake to upstream LLM endpoints from mainland China typically adds 400-600ms. Router One terminates TLS in-country and routes through provider links optimized for the China-overseas hop.
- What if I see different latency from my own machine?
- Per-request trace in the dashboard shows the latency Router One observed for your specific request. If the dashboard latency differs materially from your client-observed latency, the gap is between your client and Router One — usually local network or DNS, not the gateway.
Related
Try the same route with your own API key
Get your API key