AI tool pricing changes, summer 2026: what 15 days of pricing-page snapshots show
By The AI Tool Bible editorial teamData captured 28 Jul–11 Aug 2026; page regenerated (UTC)
We store a snapshot of each tracked AI tool's pricing page whenever an automated extraction of it changes. This study asks how often those extractions changed in a 15-day window, and how much of that is a genuine price change. The short answer: almost every tool's extracted grid changed, but most of those changes are not price changes, and the data cannot say how many are.
Key findings
- Coverage. 387 of 914 catalogue tools (42.3%) have at least one stored pricing grid, 3,030 grids in all, captured 28 Jul 2026 to 11 Aug 2026.
- The extracted grid changed for 359 of 387 tools (92.8%). The average tool has 7.8 distinct stored grids in the window and the most has 24. 28 tools have a single grid. The median time between a changed tool's first and last stored grid is 12 days.
- That is not 92.8% of vendors changing their prices. A grid is a language-model extraction of the page, and its fingerprint moves when any field moves, including feature wording, plan order or a plan being missed. Of 2,069 consecutive grid pairs where page text was stored for both, 71.3% (1,475) had byte-identical stored page text. Across all 2,643 consecutive pairs, 29.5% (779) had identical tier names and monthly and annual price strings, with only other fields differing.
- Direction of the lowest listed paid price (first versus last stored grid, tools with two or more grids): 261 the same, 22 higher, 19 lower and 57 where no price could be compared. Because extraction variation alone can move the lowest listed price (for example when a cheap tier is missed in one extraction), the higher and lower counts are an upper bound on real price moves, not a count of them.
- Shape of the grids. In each tool's latest grid, 48.8% include a free tier and 50.9% include a contact-sales or custom tier; the most common grid has 3 tiers.
Figures on this page are computed from the database when the page is rebuilt (at most daily). Frozen reference figures, generated 2026-10-06, are committed alongside the study's SQL; live figures may differ as new snapshots arrive. The live snapshot count currently matches the frozen reference.
What this data can and cannot say
- Hash change is not price change. A stored grid is added when its hash differs from the previous one, and the hash covers every extracted field.
- It is a short window. 28 Jul–11 Aug 2026 is 15 days, so it says nothing about seasonality or longer trends.
- It is not the whole catalogue. 387 of 914 tools are covered, and not at random: see the method section.
- Extraction is imperfect. It reads pages automatically, so plans can be missed, merged or duplicated, and prices can be mis-read.
- Unchanged checks were not recorded. A tool with one stored grid was checked at least once and may have been checked many times.
By category
“Grid changed” = two or more distinct stored grids. “Price text changed” = two or more distinct sets of tier names with monthly and annual price strings, ignoring features. Both are upper bounds on real change.
| Category | Tools covered | Grid changed | Price text changed |
|---|---|---|---|
| Agents | 101 | 92 (91.1%) | 80 (79.2%) |
| Writing | 64 | 60 (93.8%) | 52 (81.3%) |
| RAG | 46 | 42 (91.3%) | 37 (80.4%) |
| Image Generation | 45 | 44 (97.8%) | 39 (86.7%) |
| Coding | 38 | 34 (89.5%) | 31 (81.6%) |
| Audio | 30 | 26 (86.7%) | 23 (76.7%) |
| Video | 27 | 26 (96.3%) | 25 (92.6%) |
| Evaluation | 17 | 17 (100%) | 14 (82.4%) |
| Fine-tuning | 11 | 10 (90.9%) | 10 (90.9%) |
| MCP Servers | 8 | 8 (100%) | 8 (100%) |
Small categories (fewer than 20 tools) should not be compared with each other.
Direction of the lowest listed price
For the 359 tools with two or more stored grids: the lowest parseable monthly price in the first versus the last stored grid, in a currency present in both. “Including $0 tiers” is pinned at zero for most tools with a free plan, so “paid tiers only” is the more informative column.
| Direction | Paid tiers only | Including $0 tiers |
|---|---|---|
| Higher at last grid | 22 (6.1%) | 24 (6.7%) |
| Lower at last grid | 19 (5.3%) | 17 (4.7%) |
| Same | 261 (72.7%) | 273 (76%) |
| Could not be compared | 57 (15.9%) | 45 (12.5%) |
Looking at every stored grid rather than just the first and last: of 331 tools with a parseable paid price, 99 (29.9%) showed more than one distinct lowest paid price during the window (83 showed two values, 16 three or more). For those, the median ratio between the highest and lowest value seen was 2. Many of these will be tiers missed or added by the extraction, not price moves.
How many tiers, and how many are free
| Tiers in grid | Latest grid per tool | All stored grids |
|---|---|---|
| 1 | 11 (2.8%) | 73 (2.4%) |
| 2 | 50 (12.9%) | 395 (13%) |
| 3 | 186 (48.1%) | 1,347 (44.5%) |
| 4 | 92 (23.8%) | 806 (26.6%) |
| 5+ | 48 (12.4%) | 409 (13.5%) |
All stored grids (3,030): 1,448 (47.8%) include a free tier; 1,661 (54.8%) include a contact-sales or custom tier.Latest grid per tool (387): 189 (48.8%) include a free tier; 197 (50.9%) include a contact-sales or custom tier.
Method
How snapshots were captured
A scheduled nightly job fetched each tracked tool's pricing page (a discovered pricing URL where we had one, otherwise the tool's site with /pricing appended) and had a language model extract the plans into structured JSON: tier name, monthly and annual price strings as written, per-seat flag, seats included, and feature bullets. The job fingerprints that JSON and stores a new snapshot only if the fingerprint differs from the tool's latest snapshot. It skips tools our catalogue labels as free, and a page that could not be fetched or that yielded no plans leaves no snapshot, so covered tools are not a random sample of the catalogue. Each snapshot also keeps the first 4,000 characters of the page text where it was available. The capture window ended on 11 Aug 2026, when the extraction step began returning empty plan lists; this study does not use the replacement pipeline.
Definitions
- Covered tool: at least one stored snapshot whose extraction contains one or more plans.
- Grid changed: two or more distinct fingerprints for the tool. Because a snapshot is stored only on a fingerprint change, each stored snapshot is a distinct grid.
- Price text changed: two or more distinct sorted lists of tier name, monthly price string and annual price string, ignoring everything else.
- Lowest listed price: the lowest monthly price among strings that start with $, € or £ followed by a number. Strings that look yearly or usage-based (per year, per token, credits) and non-numeric strings (Free, Custom, Contact sales) are skipped. Currencies are never converted; first and last are compared only in a currency present in both.
- Free tier: a plan named with the word “free”, or with a monthly price of “Free” or zero. Contact-sales or custom tier: a plan named “custom”, or a price string containing contact, custom, talk to or quote.
- Consecutive pairs: each stored snapshot compared with the tool's previous one. Page text is stored truncated, so identical stored text cannot rule out a change further down the page.
Reproducibility
Every figure above is produced by SQL over two tables (pricing_snapshots and tools); the definitions above are the definitions in that SQL. The CSV is generated by the same queries. Prices are the strings vendors showed on the page at capture time and may since have changed; confirm current prices with the vendor. Scores and verification labels elsewhere on the site are unrelated to this study; see the methodology.
Download the data (CSV, CC BY 4.0)
ai-tool-pricing-changes-2026.csv: one row per covered tool (387 rows). Columns: tool, slug, category, first_capture, last_capture, snapshots, distinct_grids, first_lowest_monthly, last_lowest_monthly, direction, currency, first_lowest_paid_monthly, last_lowest_paid_monthly, direction_paid, currency_paid. Empty cells mean no price could be parsed; direction is single_snapshot where a tool has only one stored grid. Licensed under CC BY 4.0: reuse it, including commercially, with attribution. The rest of the catalogue is at /data.
How to cite
The AI Tool Bible editorial team. "AI tool pricing changes, summer 2026: a 15-day study". The AI Tool Bible, https://theaitoolbible.com/research/ai-tool-pricing-changes-2026. Data licensed CC BY 4.0. Accessed 2026-10-06.BibTeX
@misc{aitoolbible_pricing_2026,
author = {{The AI Tool Bible editorial team}},
title = {AI tool pricing changes, summer 2026: a 15-day study},
year = {2026},
url = {https://theaitoolbible.com/research/ai-tool-pricing-changes-2026},
note = {Data captured 2026-07-28 to 2026-08-11. Licensed CC BY 4.0. Accessed 2026-10-06}
}The access date is the date this page was last rebuilt; replace it with the date you retrieved the data.
More: all studies · open data and API · how we rate and verify · editorial policy