How this data is built

Confidence legend

Every figure carries a confidence level. Nothing here should be quoted as first-party fact unless marked high.

Pricing

Blended prices are approximate USD per 1M tokens for a representative open model where the provider hosts one; proprietary-only providers list a flagship model instead. Latency figures are order-of-magnitude reference points, not SLAs.

The Blended $/1M column is a simple arithmetic mean of the input and output per-million-token prices for each provider's reference model — a rough comparison anchor, not a bill estimate. Real cost depends on your input/output ratio, caching, batching, committed-use discounts and model choice. Where a provider bills by GPU-time rather than per token, the price is shown as — rather than a fabricated number.

Latency & throughput

TTFT ms (time-to-first-token) and tok/s (sustained output throughput) are order-of-magnitude reference points for a representative open model under nominal load —not SLAs and not a live benchmark. They move with model size, prompt length, region, quantization and congestion. Treat them as "which tier of fast," not exact numbers. Uptime is a reference figure, not a contractual guarantee.

Reputation composite index

reputation.score is an editorial composite index (0-100), not a first-party metric. See each site's Methodology page for the weighting model.

The 0–100 reputation score is editorial. It is a weighted composite of public signals — developer sentiment, documentation quality, track record and operational reliability — and is not a survey, not audited, and not provided by the vendors. The reviews count is the number of weighted signals behind the score, and the trend arrow is a 90-day direction. On provider pages we also normalize it to a 0–5 scale for schema.org AggregateRating; that normalization is mechanical and does not add precision.

Independence & neutrality

• No provider pays for inclusion, ranking or a badge.
• Ordering defaults to alphabetical or to the column you sort by — there is no house favorite.
• Every provider — including any we might have a relationship with — is held to the same schema and the same confidence rules.
• Corrections are welcome; figures are re-reviewed against public sources on each dated review.

Per-provider confidence

The overall confidence flag for each listing, so nothing reads as settled fact: