Hi Product Hunt 👋 I kept getting surprised by the AI APIs I build on. A price halves and I find out from the invoice. A model I call in production picks up a retirement date buried in a docs table. Vendor changelogs, where they exist at all, are prose — not something you can diff. So I started recording it instead. AI Change Watch re-reads the pricing, model, API, SDK and security pages of 15 providers every hour, diffs each against the stored snapshot, and keeps what actually changed: — The exact before → after. "Mistral Moderation input: $0.1 → $0 (−100%)", with the source link and that model's full price timeline. Numbers are lifted from the stored snapshot; an LLM only ever polishes wording, it never supplies a figure. — Shutdown dates, ahead of time. 270 recorded deprecations, 208 with a vendor-published end date, 58 still ahead. Subscribe to the .ics and each one lands in your calendar with 30- and 7-day reminders — including when a vendor moves a date. — Questions only an archive can answer. Only 66% of recorded retirements name a replacement — Anthropic 100%, Microsoft Azure 0%. And of the 24 migrations we can price today, 17 cost *more* per token, not less. The archive is also what the rest of the site is built out of: price history for 367 models and an index of the market's mean input price, model-vs-model and provider-vs-provider comparisons, and a shutdown calendar that carries each model's named successor. Free to read, no account, RSS and .ics. Nothing publishes until a person has checked it against the source — slower, but I'd rather be right than first. I'd really like to know which providers you'd want added, and whether the deprecation calendar matches how you actually plan migrations. If a number looks wrong, please say so — every page links its source, so it's checkable in one click. — Jun

AI Change Watch

Every AI API change, recorded the day it happens

Every AI API change, recorded the day it happens