Frequently Asked Questions

Platform fee: $0.0001 per request, not per LLM call. A retried request carries one platform fee, not three.
Router fee: at most 20% of each saving Meter-SPW delivers. The exact percentage is shown on your dashboard. No savings, no router fee.

Fees accrue per call and settle at the end of each billing period, not at the moment of the call. Until they settle you can audit the saving history on the dashboard and dispute any charge.

Your BYOK provider keys are encrypted at rest with AES-256-GCM and never stored in plaintext. The encryption key is never in the database or repository. Your provider keys remain usable only so Meter-SPW can route requests on your behalf.

BYOK: add keys from different providers. Meter-SPW needs keys across providers to route each request to the cheapest one.
Managed Keys: nothing to configure. Meter-SPW auto-selects the cheapest combination of models that can still deliver a quality answer.

Your model provider, directly, for every call made under your own key. Meter-SPW fees are separate and come from your SciTech credit balance.

We integrate the cheapest new models as soon as they are officially released.

Meter estimates savings using the prices published by the provider. Reported savings are an estimate, not your actual bill. Your actual price may vary depending on LLM provider's varied pricing applied at runtime.

It depends on your funding mode.
BYOK: requests continue to the model you asked for, but Meter-SPW stops routing to cheaper models. Your provider bills you as usual, and no platform fee is charged at $0.
BYOK with cost protection (opt-in): routing continues through your grace allowance, then requests stop until you add enough credit.
Managed Keys: no grace. New paid calls stop when your available credit runs out.

No. We never see or store your payment card or other payment method details. Checkout is hosted by Stripe, which holds your card details and billing address. We only store the resulting credit balance and transaction record.
These records are retained after account deletion for financial recordkeeping.

You can submit a dispute. Every charge is backed by a per-call audit record, and fees do not settle until the end of the billing period, so there is time to review. If the dispute is approved, we refund the fees paid for that routing decision in full.

A deterministic mathematical model, not an LLM classifier layer: it measures your prompt's complexity, matches it to the cheapest capable model, and routes once, in a single LLM call.
In very rare cases the cheaper model fails and Meter-SPW retries with a more capable one, up to 3 LLM calls in total. Retries never go past the model you requested. Every call is visible in the dashboard for audit.

No. A downgrade to a cheaper model is a potential outcome, not a guarantee. A cheaper model qualifies based on (a) its price, from the provider's published pricing, and (b) its estimated quality, from test samples that are constantly updated.
Both inputs change over time, so the model selected for the same prompt can change too.

Not necessarily. A routed answer may differ from the named model's own answer in wording, length, and occasionally substance. Model output is probabilistic, so two calls to the very same model can differ too.
We offer no guarantee of equivalence between a routed answer and any other answer, whether from the named model or a benchmark run.

A max-token setting is a target, not a guarantee. A model that reasons internally before answering can use more tokens on a hard prompt than a smaller, non-reasoning model would on the same prompt.
Depending on the model selected and your prompt's complexity, actual token usage can exceed the setting.

Nothing about any individual call. A benchmark score is a graded sample at one point in time; the same model, given the same prompt, can return answers of different quality and at a different price from one run to the next.
Evidence & Methodology on the product page explains how the scores were produced.