Usage & metering

Training is billed on capacity added + tokens trained (see Pricing); each finished run writes one itemized usage line. These endpoints are read-only.

GET /overview?days=30

Counters for the dashboard header.

{ "active_runs": 1, "models": 1, "datasets": 1,
  "gpu_minutes_period": 30.48, "tokens_period": 256000,
  "spend_usd_period": 0.61, "period_days": 30 }

GET /usage?days=14

Per-day rollup for the usage charts. unit is the active billing meter. Each row's cost_usd is the charged amount; gpu_minutes/tokens are recorded alongside for transparency.

{ "period_days": 14, "unit": "1M-tokens",
  "total_gpu_minutes": 30.48, "total_tokens": 1500000,
  "total_cost_usd": 25.00, "total_jobs": 2,
  "buckets": [
    { "date": "2026-06-04", "gpu_minutes": 30.48, "tokens": 1500000,
      "cost_usd": 25.00, "jobs": 2 }
  ] }

Where billing surfaces

Billing data shows up in three places, all read-only — the UI and SDK never write it:

Endpoint	Granularity	Use
GET /jobs/{id}/result	One run	cost_usd + the run's quality/compute.
GET /usage?days=N	Per-day rollup	Charts: spend, tokens, runs over time.
GET /overview?days=N	Account totals	Dashboard header counters.

Each finished run writes exactly one usage line, with the itemized breakdown that matches the pricing model — the growth, training, and (if applied) floor amounts. An estimate made before launch reconciles directly against the final line.

Fields

gpu_minutes and tokens are recorded for transparency alongside the charged cost_usd; unit reflects the meter the run was billed on (1M-tokens for training, job for contraction).