Methodology

Every listicle about AI gateways is written by a company selling one. This page exists so you can check our work instead of trusting our motives.

How prices get here

1. We read the vendor's robots.txt first. If it disallows the pricing path, we do not fetch it.
2. We fetch the public pricing page at roughly one request per second, once a week.
3. We extract numbers with their surrounding text, and attach a confidence level based on whether a billing period follows the number and whether the context looks like a price at all.
4. A human reviews the extraction before it ships. Low-confidence extractions are not published.
5. We store a checksum of the page, so a change here is traceable to a change there.

What we do not know

Stating limits plainly is the point of this page.

• JavaScript-rendered pricing is not collected automatically. Currently affects: LiteLLM, Bifrost, Cloudflare, AWS Bedrock, Unify, Flotorch. These are added by hand and may lag.
• Negotiated and enterprise pricing is invisible to us. We only see published list prices.
• We do not model your bill. Real cost depends on your traffic shape, caching, and model mix.
• Latency benchmarks are not published yet. When they are, they will be measured from a single fixed host in US-West so week-over-week numbers stay comparable, and we will say so on every chart.

Independence

• No vendor pays to be listed, ranked, or removed.
• If we ever earn affiliate revenue, it will be disclosed on the page where it applies, and it will never change a number or an ordering.
• We are not affiliated with any gateway vendor.

Corrections

If a number here is wrong, it is our error, not yours to work around. Tell us and we will fix it and note what changed.

15 gateways tracked · last collection 2026-08-30