Gateway selectionHvoy AI3 min read

A multi-model API price basket: complete quotes, not teaser prices

Compare complete input-price baskets for five selected models at the same CNY billing unit and tier, with channel evidence, quality metrics, timestamps and clear billing limits.

Current comparison scope

A valid model scope and selection policy have not yet been configured for this guide, so no shortlist is published. This does not mean no suitable gateways exist. We do not guess the current models or relax the rules to fill a list.

Basket amount is not total workload cost

The hypothetical basket uses 1 million input tokens for each of five selected models: add their unit prices for a combined 5 million input tokens, not 1 million shared across five models. It excludes output, cache reads/writes, tools, failed retries, plans and top-up discounts. It is not total workload cost or a promise of the lowest price on the entire market.

Comparability and ordering

A valid model scope and selection policy have not yet been configured for this guide, so no shortlist is published. This does not mean no suitable gateways exist. We do not guess the current models or relax the rules to fill a list.

A gateway needs qualifying quotes for all five models; one missing component removes the basket from the ranking. For each model, choose the cheapest qualifying channel in the specified tier, breaking price ties by uptime, samples, latency and stable IDs. Sort complete basket amounts ascending, then by immutable site ID. An explicit available zero quote is valid; missing, hidden or stale is never zero.

Complete baskets and model-level evidence

No qualifying candidate channels are currently available to display.

Reconcile your own usage

  1. List your actual model mix and input/output volumes first. The equal-volume basket helps compare five-model coverage, not your usage distribution. If you use only one model, compare that model's channels within a matching tier separately.

  2. Check conditions behind the quote: cache hits and misses, long-context thresholds, promotion expiry, recharge multipliers, taxes and refund limits. Record the quote time and the tier your account can actually access. An offer you cannot buy is not a budget estimate.

  3. Run a small reproducible task and reconcile request usage, successful completions and balance changes. Repeated requests add cost; cheap input does not necessarily mean a cheap completed task. If a discrepancy cannot be explained, pause expansion and request an itemized record.

Common price-comparison mistakes

Why exclude a basket missing just one model?

A sum for four models cannot be compared directly with a sum for five. We neither fill missing prices with zero nor estimate them from other gateways. Completeness prevents limited coverage from appearing cheaper.

Can different tiers share one ranking?

This guide produces one explicitly configured shared-tier basket, without cross-tier ranking. An untiered record still does not establish identical account conditions; confirm access before paying.

What if a quote is older than the article date?

Each component's quote timestamp is the price-freshness evidence. A prose revision or a new probe on another model does not refresh an old quote. An out-of-date component removes the whole basket from comparison.

Related resources