How AI prices itself
6,278 live products with a pricing model on record: how the catalog divides between free, freemium and paid, how that varies across 29 markets, and whether any of it predicts survival.
16 September 2026 · 8 min read · Data frozen 16 September 2026
Pricing in AI is discussed almost entirely through anecdote — a well-known company raising its prices, another giving something away. This report takes the whole live catalog: 6,278 products with a pricing model on record, how that divides by market, what it correlates with, and whether any of it predicts survival.
The short version
- 3,467 of 6,278 live products are freemium, 2,206 paid and 605 free.
- 64.7% offer a free tier overall, but the market range runs from 90.2% (Music) to 10.9% (Healthcare).
- Freemium products sit in markets averaging 41.7 saturation against 35.4 for paid, and carry the highest wrapper risk at 27.8. Giving it away is a symptom of the market, not a strategy within it.
- Pricing model does not predict survival. The spread across all three models is two percentage points, and it runs the opposite way to the usual advice.
Coverage
6,278 of 6,441 live products carry a pricing model; 163 do not and are excluded. Free-tier status is recorded for the same population. “Free” means no paid plan exists, “freemium” means a free tier alongside paid plans, and “paid” means no free tier of any kind — a trial with a card required counts as paid.
Finding one: freemium is the default by a wide margin
6,278 live products with a pricing model on record. Falcoscan catalog, 16 September 2026.
55% of the priced catalog is freemium. That is a striking level of convergence for a category where the marginal cost of serving a free user is not zero and scales directly with usage. The most likely explanation is inheritance: the model was imported wholesale from SaaS, where the economics that justified it do not apply here.
Finding two: an eight-fold spread by market
Percentage of live products with a known pricing model offering a free tier. Falcoscan catalog, 16 September 2026.
Music (90.2%), Coding (89.7%) and Design (83.5%) sit at one end; Healthcare (10.9%), Human Resources (26.9%) and Legal (32.5%) at the other. The dividing line is not product complexity or price point. It is whether the person who evaluates the product is the person who can approve paying for it.
Where those are the same person, the free tier is the sales process, and a product without one is not evaluated at all. Where they are different people — and especially where the evaluator cannot legally put real data into an unapproved system — a free tier generates interest that cannot convert while still costing inference on every session.
Finding three: pricing is downstream of the market
| Pricing | Products | Opportunity | Saturation | Wrapper risk | Rating |
|---|---|---|---|---|---|
| Free | 605 | 63.7 | 35.8 | 21.5 | 4.18 |
| Freemium | 3,467 | 61.6 | 41.7 | 27.8 | 4.22 |
| Paid | 2,206 | 65.3 | 35.4 | 23.7 | 3.91 |
Mean opportunity, saturation, wrapper-risk and rating across live products on each model. Falcoscan catalog, 16 September 2026.
Paid products score highest on opportunity (65.3) and lowest on saturation (35.4). Freemium products score lowest on opportunity (61.6) and highest on saturation (41.7), with the highest wrapper risk (27.8). The gap is not large in absolute terms, but it runs consistently in one direction across all four columns.
We read this as a single underlying fact seen four ways. Markets that will pay from the first day are markets that are expensive to enter; expensive to enter means fewer entrants; fewer entrants means lower saturation and higher measured opportunity; and the products there are less likely to be thin layers over a public model, because a thin layer could not have got in. The pricing decision mostly encodes a market decision made earlier.
Finding four: pricing does not predict survival
Share of tracked products on each model that no longer resolve to a working product. Falcoscan liveness sweep, 16 September 2026.
Paid products fail at 9.2%, freemium at 8.4% and free at 7.1%. The whole spread is two percentage points, and it runs opposite to the common advice that charging early imposes useful discipline.
We do not think this means free tiers are safer. The likeliest explanation is selection: a product that charges from day one is often one that must, while a product given away is frequently backed by an organisation that can afford to. The defensible conclusion is the negative one — pricing model, on its own, carries almost no information about whether a product will still exist in two years. What does carry information, per our mortality report, is the category and the buyer.
Pricing model by market
Every market, with its pricing mix and its free-tier rate. Markets are ordered by their opportunity-to-saturation gap.
| Market | Free | Freemium | Paid | Free-tier rate |
|---|---|---|---|---|
| Healthcare | 1 | 12 | 106 | 10.9% |
| Legal | 0 | 26 | 54 | 32.5% |
| 3D & AR/VR | 9 | 39 | 19 | 71.6% |
| Gaming | 5 | 35 | 17 | 70.2% |
| Real Estate | 2 | 25 | 46 | 37% |
| Research | 12 | 36 | 31 | 60.8% |
| Avatars | 1 | 48 | 17 | 74.2% |
| Music | 6 | 49 | 6 | 90.2% |
| Human Resources | 0 | 25 | 68 | 26.9% |
| Automation | 11 | 52 | 46 | 57.8% |
| Search & SEO | 3 | 47 | 18 | 73.5% |
| AI Models | 67 | 144 | 99 | 68.4% |
| Agents | 64 | 210 | 57 | 82.8% |
| Data | 45 | 174 | 70 | 75.8% |
| Security | 32 | 63 | 153 | 38.3% |
| Finance | 12 | 93 | 161 | 39.5% |
| Ecommerce | 12 | 149 | 134 | 49.8% |
| Support | 0 | 87 | 141 | 38.6% |
| Voice | 32 | 173 | 64 | 76.2% |
| Learning | 22 | 153 | 70 | 72.2% |
| Video | 11 | 222 | 60 | 79.9% |
| Sales | 2 | 133 | 139 | 49.3% |
| Productivity | 17 | 200 | 83 | 73% |
| Coding | 106 | 271 | 41 | 89.7% |
| Design | 72 | 227 | 58 | 83.5% |
| Marketing | 1 | 118 | 158 | 43% |
| Social | 8 | 176 | 80 | 70.1% |
| Image | 47 | 271 | 73 | 81.1% |
| Writing | 5 | 209 | 137 | 61.5% |
Photo: Sóc Năng Động / Pexels. Colour-graded for Falcoscan.
Citing these numbers
Every figure here is from the Falcoscan catalog as it stood on 16 September 2026, and is frozen at that reading. Later changes to the catalog will not alter this page, so a number you quote today will still say the same thing when someone checks it.
Falcoscan, “How AI prices itself”, 16 September 2026. https://falcoscan.com/reports/how-ai-prices-itself