About

Built and maintained by Jaehyun Jeon.

computetrail is an independent project that tracks the price of large language model APIs over time. Most comparison sites show only today's pricing; we keep a daily record so you can answer questions like "has this model gotten cheaper since launch?" and "which provider keeps cutting prices?".

As of today that record covers 119 consecutive daily snapshots across438 paid models, with the most recent taken on 2026-10-25, plus 10 written guides and reports built on it.

Who runs it

computetrail is built and maintained by Jaehyun Jeon, an engineer who works with LLM APIs directly — the kind of person who has to answer "what will this actually cost per month?" before shipping a feature.

The site exists because that question was annoyingly hard to answer. Provider pricing pages show today's number and nothing else; when a price changes, the old one simply disappears. There was no way to ask "has this model gotten cheaper since launch?" or "which provider keeps quietly repricing?" — because nobody was keeping the receipts. So I started keeping them: one price snapshot every day, stored and never back-filled, for every model I could track.

Everything here — the daily collector, the price histories, the comparisons andguides — is run by one person, independently, with no affiliation to or sponsorship from any model provider.

How it actually runs

There is no newsroom here, and it is worth being plain about what the operation consists of. A collector runs on a server on a fixed daily timer, fetches the current published prices, and writes one snapshot to disk. A second timer rebuilds this site from those snapshots and deploys it. Both run unattended; neither decides what is interesting.

The writing is the part that isn't automated. Analysis scripts read the stored snapshots and produce the numbers — how many models moved, on which dates, by how much. A person then reads that output and decides whether there is anything worth publishing, and writes it. Nothing is generated and posted without a human in between.

Editorial rules

These are the rules the site is run by, and we would rather they be public:

• Every figure comes from our own stored snapshots, with the date it was recorded. If we cannot source a number to the record, it does not go in.
• No word counts, no publishing calendar quotas. If a month produced nothing worth reporting, the report says so and is short — or there is no report. Filling space with general commentary is how a data site stops being one.
• Limitations get published alongside findings. Where our data can't distinguish two explanations, we say which one we can't rule out instead of choosing the more interesting story. Our methodology page lists the known weaknesses.
• We correct in public. When an updated article changes an earlier conclusion, we say what the old number was and what it is now, rather than quietly editing it.
• We measure price, not quality. This site will never tell you which model is better. It tells you what each one costs, and how that has moved.

Independence

This site is not affiliated with, sponsored by, or endorsed by any model provider. Pricing is sourced from public data; see our methodology. No provider pays for placement or sees anything before it is published, and rankings are produced mechanically from recorded prices rather than chosen.

computetrail is funded by advertising. Ads are served by a third party and are not endorsements; our privacy policy covers what that means for your data, and ourterms set out the limits of what this data is and isn't.

Contact and corrections

Questions, corrections, or a model you want tracked:wogus711929@gmail.com.

If you think a price on this site is wrong, send the model and the date and it will be checked against the stored snapshot for that day. Corrections are made to the analysis and the presentation — never to the archived snapshot itself, because an unedited record is the only thing that makes a price history worth reading.