The AI tool price barometer

How many AI tools can you actually use without paying? How many publish a real price at all? We qualify every tool in our catalogue by hand. Here is what the numbers say, across 701 tools, updated July 2026.

What the numbers say

• 346 tools, or 49 %: you start for free and pay as you scale. Fully free stays a minority (12 tools).
• 343 tools (49 %) offer no free way in at all. For a small business that means testing at your own expense.
• Only 3 % have a pricing table we could read at the source (18 tools). The rest publish a model but no numbers, or nothing at all.
• 69 % expose an API (482 tools), which is what makes a tool fit into the rest of your stack rather than sit beside it.

By category

Categories with at least 8 tools, sorted by how far you get without paying.

What the tools actually cost, table in hand

The percentages above rest on a declared pricing model. This part goes further: 18 tools whose full pricing table we read on the vendor’s own page, tier by tier. It is tedious, which is why almost nobody does it.

• The advertised price is almost never the price paid. Across the 18 tables we read, the median gap between the cheapest paid tier and the most expensive is a factor of 7. Moving up a tier is triggered by a quota (seats, credits, contacts), not by a buying decision.
• 44 % bill in US dollars. If you pay from outside the US the amount changes every month with the rate, and your bank adds 1 to 3 % on top. No pricing page mentions it.
• 50 % keep a tier with no public price. That is the norm as soon as a vendor targets large accounts: two companies of similar size can pay very different amounts for the same product.

Method

Every tool in the catalogue is qualified by hand: pricing model, ease of use, whether it exposes an API. The percentages above are recomputed straight from that data, with no sampling. The catalogue covers AI and SaaS tools relevant to small businesses; it does not claim to cover the whole market.

Pricing tables are read by hand on each vendor’s own page, never copied from a third-party comparison: half of those pages render their prices in JavaScript, and the articles that copy them contradict each other. Every row carries the date we checked it, because a pricing table goes stale in months. When a vendor publishes no price, we write that instead of guessing.