LLM cost analysis
Written pieces that work through what the pricing data actually shows, rather than restating what providers say about themselves. Every figure in them is computed from the same 60-model dataset that drives the rest of the site, verified 2026-08-16, and every claim is one you can check against the raw data.
The reference pages answer "what does this cost". These answer the harder questions: why the market looks the way it does, and which costs the rate cards leave out.
the LLM price war→
Prices collapsed at the bottom and barely moved at the top. The data on both.
hidden LLM API costs→
Six costs that do not appear on any provider's pricing page.
LLM cost per user→
The number that decides whether an AI product works at all.
self-hosting versus an API→
The break-even is a utilisation question, and it is later than people think.
Chinese LLM API pricing→
A real and measurable gap, and the four caveats on reading it.
LLM budgeting mistakes→
Five reasons the invoice beats the projection, in order of size.
How these are written
Each piece starts from a question the dataset can answer and works forward from the numbers. Where a figure appears in the text it is computed at build time from the dataset rather than typed in, so an article cannot quietly go stale after a price change while still reading as current. Where something is an opinion or a judgement call, it is written as one.
None of this is a substitute for the reference pages. If you want a rate, use the every model compared on price; if you want a number for your own workload, use a our calculator suite. These are for the questions that a table cannot answer.