knotmeter
How fast is digitalocean's serverless inference?
Tokens
per second, time to first token and uptime for every open-weight model.
method
Knotmeter sends one request per model every 15 minutes. Requests run sequentially to avoid queueing and inflating the latency it is measuring. The prompt is versioned and carries a timestamp prefix so caching cannot turn a cache hit into a fast model.
- tok/s
- Decode speed after the first token, including reasoning tokens.
- ttft
- Time from dispatch to the first token of any kind, including reasoning tokens.
- uptime
- Timeouts, 5xx and 429s all count as down.
limits
15-minute resolution is an uptime estimate, not an sla measurement, because an outage shorter than the resolution interval can be missed entirely. Closed models are excluded because digitalocean proxies rather than hosts them, so measuring them would measure someone else's infrastructure.