Live metrics
Status
Measured through our own gateway, published whether or not it flatters us.
See live metricsPer-region health
The same up / degraded / down signal the gateway itself checks before routing a request — not a separate status feed. Recomputed roughly every 5 minutes; this page re-polls every minute.
- Asia (Bangkok)Up
- SingaporeUp
- Japan (Tokyo)Up
- GlobalUp
Public metrics
Measured, not marketed
Time-to-first-token per model, measured by sending a real request through our gateway and repeated roughly every hour. This page re-polls every minute while it stays open. A model with no reading yet shows "—", never a guess.
| Model | Latency (TTFT) |
|---|---|
 Claude Haiku 4.5anthropic/claude-haiku-4.5 | — |
Claude Opus 4.6anthropic/claude-opus-4.6 | — |
Claude Sonnet 4.6anthropic/claude-sonnet-4.6 | — |
Embed 4cohere/embed-4 | — |
DeepSeek V3.2deepseek/deepseek-v3.2 | — |
DeepSeek V4.1 Flashfw-deepseek-v4.1-flash | — |
 Inklingfw-inkling | — |
MiniMax M3fw-minimax-m3 | — |
Nemotron 3 Ultrafw-nemotron-3-ultra | — |
Nemotron Lightning 3.5 30Bfw-nemotron-lightning-3.5-30b | — |
Fastest measured model first. Sorted by time-to-first-token.
What is still missing
Not on this page yet.
- Rolling uptime history per model and per region — today this page shows only the latest reading.
- Tokens per second — only time-to-first-token is shown.
In the meantime
Create a key, or read how routing and streaming work today.