knotmeter api — digitalocean serverless inference, measured

api

Everything on the board is also available as read-only JSON. All times are in milliseconds.

GET /api/v1/models: Every model on the board.
GET /api/v1/models/:id: One model, by its id (for example glm-5.3-flash).
featured: true when DigitalOcean marks the model as featured (its recommended models), otherwise false. It follows DigitalOcean's catalog and can change at any time.
?format=csv: Either endpoint as CSV: one row per model per window, with featured as the third column, after vendor.

curl https://knotmeter.digitalocean.solutions/api/v1/models/glm-5.3-flash

Example response:

{
  "generated_at": "2026-10-12T16:50:12.000Z",
  "last_sweep_at": "2026-10-12T16:45:00.000Z",
  "method": { "max_tokens": 256, "interval_min": 15, "prompt_version": "v1" },
  "models": [
    {
      "id": "glm-5.3-flash",
      "vendor": "z.ai",
      "featured": true,
      "last": { "at": "2026-10-12T16:45:00.000Z", "status": "ok", "tps": 69.4, "ttft": 612, "tpot": 14.4 },
      "windows": {
        "24h": {
          "tps":    { "p50": 40.8, "p05": 9.3 },
          "ttft":   { "p50": 695,  "p95": 2250 },
          "tpot":   { "p50": 24.5, "p95": 107.0 },
          "uptime": { "pct": 97.89, "samples": 95 }
        },
        "7d":  { ... },
        "30d": { ... }
      }
    }
  ]
}