FAQ

Frequently asked questions

Straight answers about what Tokenmeter is, how it ranks nothing by default, and how much to trust each figure.

01What is Tokenmeter?

Tokenmeter is an independent, data-first directory and comparator of AI inference providers. It currently indexes 20 providers with structured fields for pricing, latency, throughput, uptime, OpenAI-compatibility, compliance and an editorial reputation index. It is a reference, not a marketplace or reseller.

02Is Tokenmeter affiliated with any of the providers it lists?

No. Tokenmeter is editorial and unaffiliated. No provider pays for inclusion, ranking or placement. Every listing follows the same schema and the same confidence rules, with no house favorite.

03How are providers ranked?

There is no fixed ranking. The directory defaults to alphabetical order and the comparison table sorts by whichever column you click. The reputation index is one signal among many, not an overall verdict.

04Where do the pricing figures come from?

The blended $/1M figure is the mean of input and output per-million-token prices for each provider's reference model, drawn from public pricing pages. It is a comparison anchor, not a bill estimate — real cost depends on your token mix, caching, batching and committed-use discounts. Providers billed by GPU-time show "—" instead of a per-token price.

05Are the latency and throughput numbers benchmarks?

No. TTFT (time-to-first-token) and throughput (tok/s) are order-of-magnitude reference points for a representative model under nominal load, not live benchmarks or SLAs. They shift with model size, prompt length, region and congestion.

06What is the reputation score, exactly?

It is an editorial 0–100 composite index of public signals — developer sentiment, documentation quality, track record and reliability. It is not a survey, not audited, and not supplied by the vendors. On provider pages it is also normalized to a 0–5 scale for structured data.

07What do the confidence badges mean?

Each figure is tagged high, medium or seed. High means it was confirmed against the provider's public docs on the review date; medium means it is public but not re-confirmed on that date; seed means it is illustrative and pending first-party verification — not to be quoted as fact. The dataset was last reviewed 2026-07-10.

08Should I trust these numbers for a purchasing decision?

Use Tokenmeter to shortlist and compare, then verify the exact figures against each provider before you commit. Seed figures in particular are illustrative. See the Methodology page for the full sourcing model.

09How often is the data updated?

The dataset carries a dated last-reviewed field (currently 2026-07-10) and a schema version. Figures are re-checked against public sources on each dated review.