1NFER exact-model retail billing policy, version 3, 2026-10-05

This is 1NFER's own retail policy. It does not describe or certify the
upstream supplier's accounting, procurement cost, model weights or profit.
The working API origin is https://1nfer.net/v1.
This version adds $1 test funding at reference rates. Earlier policy versions
and already issued invoice quotes/request receipts are preserved.

For each accepted exact model identifier, retail USD charges use the validated
reported prompt_tokens and completion_tokens, separately, multiplied by the
current advertised input/output USD rates per 1,000,000 tokens. Charges round
to eight decimal USD places. total_tokens is advisory: a mismatch is recorded
as a receipt anomaly and does not supply a charge unit or an invented count.
Cached prompt tokens and cache writes are included in prompt_tokens, with no
additional retail cache fee and no cache rebate. Reasoning tokens reported as
a subset of completion_tokens are included with no additional retail fee.
Missing, negative, malformed or incoherent usage is not estimated: the request
remains under review with its authorization hold. The supplier's USD cost is
unknown and is shown as null, not zero. This policy does not quote the official
provider's cache-specific pricing, Batch, search/tool fees or geography taxes.

Reference: https://platform.claude.com/docs/en/about-claude/pricing
Reference input/output USD per 1,000,000 tokens (standard global synchronous):
claude-sonnet-4-6: 3 / 15; claude-sonnet-5: 2 / 10;
claude-opus-4-6, claude-opus-4-7, claude-opus-4-8, claude-opus-5: 5 / 25;
claude-opus-5-5: 4 / 20; claude-fable-5: 10 / 50.
Identifiers pass through unchanged. No silent alternate model or provider is
used. Matching an identifier does not independently attest supplier weights.

The public default and accounts without an earned discount use 100% of each
model's reference input/output rate. Minimum API credit is $1; service fee is $0.
A confirmed credit below $10 grants no launch discount.
An account earns the launch factor from its highest SINGLE confirmed API-credit
invoice: $10 to below $20 -> 50%; $20 to below $100 -> 40%; $100 to below $300 -> 30%;
$300 to below $500 -> 25%; $500 or more -> 20% of the reference rates.
The credit amount excludes fees. Smaller cumulative payments do not combine
to qualify. Gifts, manual credits, setup/business payments, legacy Telegram
credits and pending invoices do not qualify. Verified sponsor funding of the
same API-credit invoice qualifies for its bound recipient account.
Once earned, the highest single-credit entitlement persists; spending balance
or receiving a later smaller credit does not remove it. Key rotation preserves
the account. The tier multiplier persists; official-reference price changes
can affect future calls, not a previously reserved or settled receipt. Each
request freezes its own reference, tier and retail rates at reservation.
Invoice quotes are snapshots; a quoted future tier applies only after verified
invoice confirmation. New crypto invoices are available only when live payment terms report enabled.
The native invoice shows exact asset/network/amount and a 120-minute window.
After expiry, existing invoices can still be checked for late confirmation
within the stated recovery window; do not initiate a new transfer after expiry.
No registration, free trial, referral count or gift grants a paid-tier upgrade.

Current prepared route limits: 4096 context units, 1024 output tokens, 16384
request bytes. Text content only. Standard function tools are verified only for
claude-sonnet-4-6; other seven models reject tools. Provider-managed search/tools,
images and other modalities are not supported by this release. Streaming is
buffered: completion validation and settlement precede the returned SSE frames.
Each attempt reserves a maximum authorization of $0.10 from available account
credit; an optional lower max_charge_usd can reduce that ceiling. A hold is not
a fixed fee or an estimated bill. If no verified buffered output is delivered,
the customer charge is waived and its hold is released; the supplier/billing
review record is retained. The identical unconfirmed payload and any recorded
idempotency key cannot be replayed automatically. Separate new tasks can use
remaining credit. No automatic retry or model substitution occurs. The account must have sufficient unreserved credit. A confirmed merchant payment and the matching idempotent credit ledger are
required before API credit becomes usable. Creating an invoice does not pay it;
no wallet transfer is automatic. BTC/LTC/XMR and conditionally sender-bound
ETH/TRX/USDT-TRC20 are supported. Live settlement testing is in progress. The separate
hs-mini local free demo keeps its existing quality and quota limitations.