About SelfBench
A small, deliberately boring tool: published LLM prices, in one place, with the sources attached.
Model pricing is public information that is genuinely hard to use. Vendors publish rates on their own pages in their own formats, change them without notice, and never make it easy to compare one request against another. The result is that most teams pick a default model on vibes and discover the cost later.
SelfBench exists to fix the small part of that: take a request size and a traffic estimate, price it across the models you are considering, and show where the numbers came from. Every model page links to the vendor's own pricing page and the date we verified it.
What this site is not
- Not a benchmark. We do not publish quality scores. A per-token price says nothing about whether a model will do your job.
- Not a proxy. We do not sit between you and your provider. No prompts pass through here.
- Not a broker. We have no affiliate deal with any vendor, so no model is promoted for a commission.
How it is run
SelfBench is built and maintained by one person, with automated checks that fail the build if the dataset is missing or malformed. The Pro plan exists to cover the cost of keeping the dataset current — the free tools are not a demo, they are the point.
Corrections, model requests and complaints are all welcome: contact us.