System status

Components

• API gateway

  apis.opengateway.ai — OpenAI-compatible endpoints

  Operational

• Control plane

  Accounts, API keys, billing

  Operational

• Web console

  opengateway.ai dashboard and docs

  Operational

• deepseek/deepseek-v4.1-flash-ultrafast

  UltraFast

  Operational

  TTFT p50

  417 ms

  TTFT p90

  905 ms

  Output speed

  454 tok/s

  24h peak 1490 tok/s

  Success rate

  99.89%

  TTFT p50 · 24h · 340 ms–554 ms

  Output tok/s · 24h · 447–892

• z-ai/glm-5.3-flash-ultrafast

  UltraFast

  Operational

  TTFT p50

  665 ms

  TTFT p90

  942 ms

  Output speed

  394 tok/s

  24h peak 797 tok/s

  Success rate

  100.00%

  TTFT p50 · 24h · 192 ms–1.73 s

  Output tok/s · 24h · 192–471

How we measure

• Every number is measured at the OpenGateway API gateway on real customer traffic over the last 30 minutes, not on synthetic probes.
• TTFT (time to first token) is the time from the gateway sending a streaming request upstream to receiving the first chunk. It includes prefix-cache hits, so repeated prompts are faster than cold ones. p50/p90 are interpolated from latency histogram buckets.
• Output speed is output tokens of successful requests divided by their generation time (request duration minus TTFT), weighted by tokens. Non-streaming requests have no TTFT, which makes this figure slightly conservative.
• 24h peak is the highest output speed of any 5-minute window in the last 24 hours, computed the same way. Windows with fewer than 100 successful requests are skipped so a handful of requests cannot set the peak. It is a window aggregate, not the fastest single request.
• Success rate excludes client errors (4xx). Status is Operational at 99% or more, Degraded at 95–99%, Outage below 95%. With fewer than 30 requests in the window we show "Measuring" instead of a number.
• The 24-hour charts use one-hour buckets; empty gaps are hours with too little traffic.