Fetching the latest probe
24-hour samples—
Success rate—
First text P50—
First text P95—
| Time (SGT) | Model | Result | First text | Total | token/s* |
|---|---|---|---|---|---|
| Loading the live table. A readable published snapshot is provided below. | |||||
Published snapshot
| Time (UTC) | Model | Result | First text |
|---|---|---|---|
| gemini-3.5-flash-lite | Passed | 2.31 s | |
| gemini-3.5-flash-lite | Passed | 2.40 s | |
| gemini-3.5-flash-lite | Passed | 1.97 s | |
| gemini-3.5-flash-lite | Passed | 2.07 s | |
| gemini-3.5-flash-lite | Passed | 2.29 s | |
| gemini-3.5-flash-lite | HTTP_500 | — |
Read the measurements
- Probe route: NAS probe → configured egress proxy → public token.subfor.me endpoint → gateway → model. This is not an independent overseas monitoring location.
- First text: Time from sending the request to the first non-empty text fragment, including connection, network and model wait. SSE heartbeats are excluded.
- token/s*: Final upstream completion_tokens, minus reported reasoning_tokens, divided by the interval between first and last visible text fragments. This is approximate throughput, not per-token timestamps. A single fragment or interval under 0.1 seconds is not reported.
- Samples and success: All executed checks in the last 24 hours count, including failures. P50/P95 use successful samples only. Small samples do not establish stable performance.
- Budget and freshness: The probe key has a finite allowance and expiry. Generation stops when credits run out or authentication fails. Data older than 40 minutes is marked stale.
Model names are declared by the provider. This page measures connectivity and performance; it is not model identity or capability certification.