Frequently asked questions

nfer is an AI model deployment cost comparator. Once you've picked a model - Llama 3, Mistral, Qwen, Claude, or another - nfer shows you every place it can run (token-priced API, dedicated/reserved instance, or rented GPU) and tells you which is cheapest for your specific workload, across several providers worldwide.

You search for a model, set assumptions about your workload (monthly token volume, input/output ratio, cache hit rate, target utilization, quantization), and nfer normalizes every provider's pricing to a single monthly-cost figure. Results are sortable and filterable by sovereignty, certifications, region, license, and pricing type.

nfer compares the major deployment paths for AI models: API (per-token, rate set by the provider), dedicated/reserved capacity (a committed instance billed hourly with a fixed throughput ceiling - better economics at sustained volume), GPU rent (hourly hardware like an A100 or H100 you operate yourself), co-location (your hardware in a third-party data centre), and self-hosting (your hardware on your own premises). All are normalized to monthly cost on the same model so they compare like-for-like.

Yes - select more than one model in the search bar and the comparator shows results for each side-by-side. A dedicated A/B view that lines two models up directly is on the roadmap.

A GPU's raw price is hourly, not per-token. nfer estimates a blended per-million-token rate by combining the GPU's hourly rate with realistic throughput assumptions (model size, quantization, batch size, target utilization). The number is an estimate, not a guarantee - adjust the assumptions on the home page to see how the blended figure changes for your workload.

Two common reasons: (1) for GPU and reserved capacity, provider pages quote hourly rates while nfer translates them to a workload-blended per-token figure for like-for-like comparison; (2) providers frequently run promotional credits or volume discounts that aren't in their public price list. If a price looks wrong, please send the provider URL - we'll investigate as soon as possible.

Prices are synced from each provider's public pricing source. You can find the latest update time on every model card in the comparator.

Public, provider-published pricing pages and API price lists. We do not negotiate private pricing.

We pull prices directly from each provider's public pricing pages and aim to match them faithfully. Reserved/committed pricing is approximated when providers don't publish a clear per-hour equivalent - in those cases the page says so explicitly. Always verify the headline number against the provider's own page before signing a contract.

Coverage expands continuously. If something you need is missing, email [email protected] with the provider and model - most additions take under a day once we have a stable pricing source URL.

Not yet via a public API. A paid price-tracker API is on the roadmap for teams that need programmatic access. Until it ships, email [email protected] if you need a one-off export.

nfer is built by M3T, an AI company building distinctive products. Email [email protected] or book a free 30-minute consultation via the link on the home page. We aim to reply to every message quickly.