Fronset — what it costs and how to start

Before you choose a plan: on the free and shared_anon plans, a prompt template you store with us may be adopted into our shared library and republished, word for word. Only private keeps templates yours. Keep commercially sensitive templates off the free and shared_anon plans, and upgrade before your first call. Exactly what applies, and when

Two ways to pay for the models

Bring your own keys. You vault your provider credentials; your provider bills you directly for the tokens — comparison calls included — and we charge only the per-call fee.

Let us manage them. We serve the call on our provider accounts, drawn from prepaid credit you load in advance. Model usage is billed at the provider’s own rates while one of our tasks is still being measured, with every model call behind your request on your invoice at cost, and at the task’s published rate once one of our tasks has settled — a rate set from measured cost, with no markup, corrected in both directions as costs move. A task of your own is always billed at the provider’s rates: we never put it on a published rate. Buying credit costs 5.5% of the purchase (minimum $0.80).

The per-call fee

The same on every plan, whichever way you pay for the models. Counted per request, not per model attempt — a call that fans out to several models to pick a winner is one call. Why the fee is the same whichever model answers, and where our incentives point: how routing works.

Calls per month	Per 1,000 calls
First 10,000	included
10,000 – 50,000	$2.00
50,000 – 250,000	$1.50
250,000+	$1.00

Graduated: each band applies only to the calls inside it.

What this comes to on real traffic: one month of our internal customer’s calls, every fee included, against the same calls on one model. Read the case study →

Plans

Free — $0, paid for with data. We keep the prompts you send, word for word, may re-run them on new models, and pool the quality grades from your calls on our tasks with other customers’. Pooling can’t be undone. Send no personal data on it.

shared_anon — $49 a month or $490 a year. Keeps your prompt text off the call record. A batch keeps a working copy while it runs, and grades from your calls on our tasks are still pooled. Stored templates may be adopted into our shared library and republished. Subscribe from the Billing page in the console, or with a management key through checkout. Switch between monthly and annual, change your card, or cancel, from Manage subscription on the Billing page: monthly to annual applies at once, with credit for the rest of the month; annual to monthly and cancelling apply at the end of the paid period.

private — by conversation. Grades stay in your account, and templates are never republished. We sell private by conversation: write to us.

Upgrade before your first call. Each call is stored under the plan in force when it runs, and pooled grades can’t be withdrawn afterwards.

Limits

Every limit is counted per account: all the keys you hold share it.

	Free	`shared_anon`	`private`
Inference requests per second	10	20	60
Management writes per hour	300	300	300
Calls or spend per day or month	no cap	no cap	no cap

Credit purchases by card carry a monthly ceiling per account, which bounds payment exposure while an account is new — it is not a limit on usage. Yours is shown on the Billing page of the console; to have it raised, write to support.

Availability and support

Tier	Channel	First response	Uptime	Status page
Free	Docs and email, best effort	none	none	public
Shared-anon	Email	Next business day	Published 99.5% monthly target	public
Private	Priority email, named contact	6 hours	Published 99.5% monthly target, plus incident communications and postmortems	public, plus incident subscription

First-response times are measured during support hours: 08:00–20:00 Pacific Time, Monday to Friday, excluding US public holidays. The uptime figure is a target, not a warranty: no plan carries service credits. Live status

What Fronset is not

Not a chat API. No streaming, no conversations (one user message per call), no tool or function calling. Requests that need them get a clear error; we don’t pretend.
Not for images or audio. Text in; text or JSON out.
Not a place to tune sampling. Temperature, max_tokens, seed and the rest are accepted and ignored. The task owns its settings, and every response lists what we ignored.
Not a model you pick call by call. You name the task. On our tasks we choose the model: pinning and a confidence floor are refused, whichever keys serve it, and once a task is settled the bar is ours and we publish it with the rate. On a task of your own you set the bar and can pin a model (pinning turns off failover).
Not for EU or UK personal data. The service isn’t offered for personal data subject to the EU or UK GDPR.
Not free of data terms on the free plan. On the free plan we keep the prompts you send word for word, may re-run them on new models on our provider accounts, and pool the quality grades from your calls on our tasks with other customers’. Pooling can’t be undone. Send no personal data on it. shared_anon keeps your prompt text off the call record, though a batch holds a working copy while it runs and its grades on our tasks are pooled too; private keeps grades to your account. Upgrade before your first call: each call is stored under the plan in force when it runs (details, including batch).
Not an SLA with credits. On the paid plans we target 99.5% monthly uptime; the free plan has no uptime target. No plan carries service credits.
Not self-serve for private yet. We sell private by conversation.
Not a leaderboard of the smartest model. We rank the models that are good enough for a task by what they cost to run it.

Questions we expect →

Getting started

Self-service signup is not open yet — we are opening in small waves. Request an invitation and we will send you one when it is your turn. This is not the benchmark journal; joining it does not subscribe you to anything else.

You will get two keys: one for inference, one for managing the account. They are shown once.

What you need before your first completion depends on how the task is served. A task you create yourself runs on your own provider keys unless you ask otherwise, so vault one and no credit is required — your provider bills you directly. A task served on our keys draws prepaid credit for the model usage, and a call made before you load any is refused. Either way there is a per-call fee, and every account has a monthly allowance of calls that carry none.

Listing the models available to you works immediately in both cases, so that is the quickest way to check your key is live.