Forecasting Pricing

Forecasts are billed in prepaid InertialAI credits — one balance covers every model, so you never reconcile multiple invoices. Each request returns exactly what it cost, and the same value is set on response headers, so you can track spend per call.

Every forecast is billed on the same two components — a flat request fee and a per-point rate — and a reasoning request adds one more, usage-based component on top:

Component	Applies to	Rate
Request fee	every forecast	1 credit / request
Data point	every forecast	0.03 credits / point
Reasoning usage	reasoning calls that use an LLM	1.15 credits per $0.01 of model usage for that call

Forecast-only

A plain inertialai-forecast request is deterministic — you can compute the cost before you send it. A "point" is one quantile at one horizon step:

points  = horizon × number_of_quantiles
credits = 1 + points × 0.03

The length of your series does not change the price; only the size of the forecast you request does.

Example. A 14-step forecast with the default three quantiles (0.1, 0.5, 0.9):

points  = 14 × 3 = 42
credits = 1 + 42 × 0.03 = 2.26

The reasoning version

inertialai-forecast (reasoning) is billed as the same request fee and per-point rate as above, plus a reasoning charge for calls that actually invoke an LLM. There's no separate reasoning base fee or per-token rate — a reasoning request costs exactly what a forecast-only request of the same size would, plus this one extra line item.

A request sent without context or image_urls self-reasons directly from the series (clipping, rounding, widening intervals) with no LLM call, so no reasoning charge applies — you pay the same as forecast-only. The reasoning charge only appears once you supply context and/or image_urls, and it scales with the model you pick: the reasoning model catalog sorts models from small to large, so picking a smaller model keeps this line item small too.

Example (self-reasoning, no context). Series of 8 points, horizon 7, default quantiles, no context/image_urls:

points  = 7 × 3 = 21
credits = 1 + 21 × 0.03 = 1.63   # identical to a forecast-only request this size

Example (with context, using a reasoning model). Horizon 4 with default quantiles, plus context describing a promotion:

points  = 4 × 3 = 12
credits = (1 + 12 × 0.03) + reasoning usage
        = 1.36 + 1.15   # this call used $0.01 of the selected model
        = 2.51

Automatic routing

You do not choose the price model directly — it follows what you send:

• Send inertialai-forecast with no context and no image_urls → billed as forecast-only.
• Send any context text or image_urls → the request is automatically upgraded to the reasoning version and billed as such, even if you asked for inertialai-forecast.
• Send inertialai-forecast (reasoning) with no context → still routed through reasoning, but since no LLM call runs, you're billed the same request + data-point charge as forecast-only, with no reasoning line item.

The model field in the response tells you which model actually ran, so you can always confirm how a request was priced.

Seeing what a request cost

Every response includes a usage object and two headers. usage.credits is always the exact amount charged; usage.total_tokens is reported for visibility into the reasoning call and does not itself determine the price:

"usage": {
  "prompt_tokens": 22,
  "completion_tokens": 0,
  "total_tokens": 22,
  "credits": 2.26
}

• x-api-total-request-token-count — total_tokens
• x-api-total-request-credits-used — credits

A request is only charged against your balance after it succeeds. If you do not have enough credits, the request is rejected before any forecast work is done.

Choosing the cheaper path

• For recurring, high-volume jobs where history is enough, use inertialai-forecast — the per-point price is small and fully predictable.
• Add context only when it matters — each reasoning call that uses an LLM adds a usage-based line item on top of the same request + data-point charge, so batch or cache where you can, and pick a smaller model from the catalog when a lighter one will do.

Rates shown are the current defaults and may be configured per account. See the platform pricing page for the latest published prices.