AI Price Watch 2026, the dataset behind the AI pricing pages
14 files recording what AI providers say on their own pages about prices, plan tiers, scheduled price changes and model retirements. Read on 26 October 2026, re-read weekly, with the SHA-256 of every saved provider page. Provider-stated files are kept apart from derived files and from one unverified file.
What is in the dataset
14 files, listed with their kind, size and SHA-256. A changed hash means the file changed. The same list is machine-readable in data.json.
Each parsed by pattern from a saved copy of a provider page, with the page hash in the evidence manifest. The build stops if a page stops matching.
Computed from public sources (LMArena ratings, OpenRouter prices) and labelled as derived on every page that uses them.
A month-by-month change log from a community price file. Several apparent rises were stale records, so it is kept apart and labelled.
Saved copies read on 26 October 2026, each with a SHA-256 hash. The copies are not redistributed; the hashes identify what was read.
| File and hash | What it holds | Size |
|---|---|---|
scheduled-changes.jsonsha256 7862b9177918 | Provider-stated. Provider-stated dated price changes (Google Gemini 4 Argon, Gemini 3.8 and 3.6 Flash) and dated statements for DeepSeek and OpenAI. | 3.6 KB |
tier-structure.jsonsha256 9140df569683 | Provider-stated. OpenAI Standard, Batch, Fast and Ultrafast prices; Anthropic prices, cache multipliers and the stated tokenizer note. | 7.4 KB |
deepseek-pricing.jsonsha256 06d181e4ef7c | Provider-stated. DeepSeek peak and off-peak prices, peak hours in UTC, and search and fetch billing from 2026-11-01. | 8.3 KB |
google-pricing.jsonsha256 5fc2eab6c302 | Provider-stated. Gemini API text-model prices, free-tier availability and the data-use row of the pricing page. | 6.8 KB |
search-tool-prices.jsonsha256 49526068d8cf | Provider-stated. Web search tool price per 1,000 calls for DeepSeek, OpenAI, Anthropic and Google. | 3.1 KB |
tier-differentiation.jsonsha256 25ec7e42bafe | Provider-stated. ChatGPT, Claude, Gemini and Mistral consumer plans as displayed on the read date (US), with stated usage multiples. | 5.5 KB |
plan-ladder.jsonsha256 b6a443f58cac | Provider-stated. Hand-sourced plan ladder behind tier-differentiation.json; ChatGPT dollar figures were read from a screenshot and are not raw-archived. | 3.0 KB |
retirements.jsonsha256 b45431aa96ad | Provider-stated. Model retirements and notice periods parsed from the Anthropic, OpenAI and Google deprecation pages, with upcoming retirements and Anthropic migration price change. | 65 KB |
mortality-ledger.jsonsha256 2d249258a217 | Provider-stated. Groq stated shutdown and announcement dates joined to our own probe dates for the Groq models on our audit panel, and how long each named replacement lasted. One provider, one small panel. | 17 KB |
price-per-capability.jsonsha256 d5584c1f4a38 | Derived, labelled. LMArena overall text rating (CC BY 4.0, LMArena) joined to OpenRouter list prices. A human-preference rating, not accuracy. | 52 KB |
arena-overall-2026-10-22.jsonsha256 09bd0b845fbd | Derived, labelled. The LMArena overall rating table used by price-per-capability.json. | 123 KB |
dispersion.jsonsha256 4b67f9f233f3 | Derived, labelled. What each OpenRouter host charges for the same open-weights model. Hosts differ in quantisation, context, speed and data policy; a gap is not evidence of overcharging. | 166 KB |
ai-price-calendar.icssha256 9be8da11a0a9 | Provider-stated. Subscribable calendar of dated price changes and upcoming retirements (iCalendar). Earliest-possible retirement dates are marked TENTATIVE. | 35 KB |
candidates-unverified.jsonsha256 857555fb58ee | Unverified. UNVERIFIED. Month-by-month changes in a community price file (LiteLLM). Several apparent rises were stale records, retired names or peak prices. Do not cite any row as a provider price change. | 92 KB |
Subscribe to the dated events in your calendar app: webcal link. Earliest-possible retirement dates are marked tentative.
The pages that read it
Each page below is built from these files, shows what the provider page says, and names the page it was read from.
- Every dated AI price change and model retirement
- Which AI plan gets the flagship model
- OpenAI API pricing: Standard, Batch, Fast and Ultrafast
- Gemini API pricing and the 1 December change
- DeepSeek API pricing, peak hours and the search fee
- AI web search tool prices per 1,000 calls
- Cheapest AI model for your task
- How much notice providers give before retiring a model
- Groq: stated shutdown dates against what we observed
How to cite it
Couey, V. W. (2026). AI Price Watch 2026: what AI providers state about prices, tiers, scheduled changes and model retirements. Nesyona. https://nesyona.com/research/ai-price-watch-2026/ (data read 26 October 2026). https://doi.org/10.5281/zenodo.23664178 (version 1, 26 October 2026). To cite every version: https://doi.org/10.5281/zenodo.23664177, which always resolves to the latest.
Cite the file and its SHA-256 for a specific figure. Because the files are rebuilt weekly, the read date and hash identify the version.
How it is built, and what it does not cover
- One storefront (United States), the providers named on each page, and the read dates shown. A provider not listed may have changes we did not see.
- List prices only. Tokenizer changes and hidden reasoning tokens change what a task costs without changing the list price, and are not measured.
- Plan pages (ChatGPT, Claude, Gemini, Mistral) were read from rendered text, and the ChatGPT dollar figures from a screenshot; they are not archived as raw HTML.
- No provider was asked to comment before publication. A provider or reader who finds a wrong figure can tell us; we correct it and date the correction.
FAQ
What is AI Price Watch?
A dated set of files recording what AI providers state on their own pages about prices, plan tiers, scheduled price changes and model retirements, plus a few derived files built from public sources. It was read on 26 October 2026 and is re-read weekly.
Can I reuse it?
Yes. The files are published under CC BY 4.0: reuse is free with attribution to Vincent Wesley Couey and a link to this page. LMArena ratings inside the derived files are CC BY 4.0, attributed to LMArena.
How do I know a number is what the provider said?
Each provider-stated figure was parsed by pattern from a saved copy of the provider page, and the SHA-256 of that copy is recorded. The build refuses to write a file if a page stops matching its expected structure. A reader can fetch the provider page and compare.
Why is one file marked unverified?
candidates-unverified.json lists changes in a community-maintained price file between monthly snapshots. Checked against provider pages, several apparent price rises were stale records, retired model names or peak-hour prices. It is kept so the check is visible, and it should not be cited as a list of provider price changes.
Does it have a DOI?
Yes. Version 1 is https://doi.org/10.5281/zenodo.23664178, deposited on Zenodo on 26 October 2026 with the provider-stated files, the README and the evidence manifest. The DOI for all versions is https://doi.org/10.5281/zenodo.23664177. Weekly re-reads on this page are not separate DOIs; they become new versions when the data has changed enough to deposit again.