Research report · ai-agents
AI Agent Pricing Models: 21 of 93 Tools Bill by Usage
We classified the pricing meter on 93 of the 94 AI agents published in our index on 26 August 2026. 21 bill by usage, 15 of them voice or support agents.
On this page
Pricing model across the AI agent index
Of the 93 AI agent listings with a classified pricing model, of the 94 published on 26 August 2026, the largest group is “Freemium” at 45 (48%). Measured 2026-08-26.
Of the 93 AI agents we classified in our index on 26 August 2026, 21 bill by usage, 45 are freemium, 17 are flat subscriptions and 10 are free. The classification records what each vendor’s own pricing page uses as its headline meter, which is the unit the bill scales with.
Correction, 11 September 2026. This report originally read “AI Agent Pricing Models: 20 of 93 Tools Bill by Usage” and printed the breakdown “44 are freemium, 20 bill by usage, 19 are flat subscriptions and 10 are free” out of a population of 93. No collector run reproduces those four counts together, and no run in this index’s history holds a published population of 93 at all. It stood at 92 on 18 August and 94 on 19 August, so the 93 could only be reached by dropping one listing from the 94 without saying which. This edition is restated to the reading that does reproduce, collector run 200142 of 26 August 2026: 94 listings published, 93 of them carrying a pricing model, and the excluded one named below. Three tools the old text placed by name have moved with it: Auto-Respond is a flat subscription rather than usage-based, Devin is freemium rather than usage-based, and Claude Code is freemium rather than a flat subscription. If you cited the 20-of-93 headline or the 44/20/19/10 split, this is the correction.
What this counts, and what it leaves out. On 26 August 2026 this index held 94 published listings. 93 of them carry a pricing-tiers table we classified into one of four shapes: free, freemium, flat subscription or usage-based. Every count on this page is out of those 93. The one exclusion is Agentix Labs, the single listing in our agencies category: it sells services rather than a product with a published price list, so there is no pricing table to classify, and it is left out rather than counted as free.
The distribution is not the useful part. The useful part is that the usage meter is not spread across the index at all: it sits almost entirely in the agents that talk to your customers, and almost nowhere in the agents that sit beside your developers.
The headline count, out of 93
21 of the 93 AI agents classified in our index on 26 August 2026 bill by usage, where the headline meter is a unit of work rather than a seat. The shape was determined for every one of those 93, so each count below is out of the whole classified set and there is no unknown bucket to set aside.
- 45 of 93 (48%) are freemium, with a real free tier or a free open-source core beneath the paid tiers
- 21 of 93 (23%) are usage-based, where the headline meter is a unit of work: a minute, a resolution, an action, a credit
- 17 of 93 (18%) are subscription, a flat recurring price per seat or per plan with no free tier
- 10 of 93 (11%) are free, with no paid tier of any kind
The meter follows who the agent talks to
Splitting the index by category shows where each meter lives. The row totals sum to 96 rather than 93 because three listings, Sierra, Botsify and ElevenLabs Conversational AI, sit in two categories each and are counted once in both. Sierra and ElevenLabs are usage-based and Botsify is a subscription, which is why the usage-based column sums to 23 and the subscription column to 18.
| Category | Freemium | Usage-based | Subscription | Free | Total |
|---|---|---|---|---|---|
| Coding agents | 17 | 1 | 3 | 4 | 25 |
| Agent platforms | 8 | 3 | 4 | 1 | 16 |
| Agent frameworks | 10 | 0 | 0 | 4 | 14 |
| Voice agents | 2 | 8 | 2 | 0 | 12 |
| Research agents | 5 | 4 | 1 | 1 | 11 |
| Support agents | 0 | 7 | 2 | 0 | 9 |
| Sales & marketing agents | 2 | 0 | 5 | 0 | 7 |
| Agent tools | 1 | 0 | 1 | 0 | 2 |
Voice agents and support agents hold 15 of the 21 usage-based slots between them. Frameworks hold none at all.
Customer-facing agents meter, developer-facing agents do not
Take the categories where the agent handles an end customer, meaning voice agents and support agents. No tool sits in both, so that is 21 distinct listings. 15 of the 21 voice and support agents in our index bill by usage, which is where the usage meter concentrates.
All seven usage-based support agents meter a resolved conversation or a handled ticket: Sierra, Fin, Decagon, Crescendo, Cresta, Ada and Zendesk AI Agents. Fin is billed per resolution on Intercom’s own pricing page. Auto-Respond was an eighth in the edition this page previously printed; on this reading its pricing table is a flat monthly price, so it sits in the subscription column instead.
All eight usage-based voice agents meter a minute of call time: Vapi, Retell AI, Bland AI, PolyAI, Synthflow AI, ElevenLabs Conversational AI, Deepgram Voice Agent API and the OpenAI Realtime API. The one voice agent that joined the index since the 18 August edition, VoiceAgent, is not among them: its $12.50/month Starter plan is a flat price with included minutes, not a per-minute meter, which makes it this category’s second subscription tool alongside SkipCalls.
Now take the categories where the agent works alongside an employee, meaning coding agents and frameworks. That is 39 distinct listings with no overlap. 1 of the 39 coding agents and frameworks in our index bills by usage, Muse Code, against 15 of 21 among voice and support agents. Muse Code is that one. Devin was the second here on both the 18 August edition and the edition this page previously printed; on this reading its pricing table classifies as freemium. Instead 35 of the 39 are free or freemium, and the remaining three are flat subscriptions: Factory, Augment Code and Gemini Code Assist. Claude Code was a fourth on the previous edition and classifies as freemium here.
15 of 21 against 1 of 39 is not a gradient. It is the most useful thing on this page for anyone building a budget, because the two shapes fail in opposite directions. A voice or support agent is priced like infrastructure, so the bill scales with how much work it does, which makes it near-free in a pilot and hard to forecast at volume. A coding agent or framework is priced like a tool: free to start, then a flat rate per developer, largely indifferent to how hard the team leans on it.
The reason is visible in what each category can meter. A resolved support ticket and a minute of phone call are units the buyer already counts and already has a human cost for, so a price per unit can be compared against something. There is no equivalent unit for “wrote some code,” because the industry has no agreed meter for developer output, and the seat survives as the fallback.
Open source and metered billing do not mix
Crossing the pricing field against the open-source field gives the cleanest separation in this index. 93 of the 94 published listings in our index have a determined open-source status, and the 94th is the agency listing this census excludes. So this table covers the same 93 listings as every count above, and there is no undetermined column to set aside.
| Open source | Closed source | Total | |
|---|---|---|---|
| Free | 10 | 0 | 10 |
| Freemium | 23 | 22 | 45 |
| Subscription | 0 | 17 | 17 |
| Usage-based | 0 | 21 | 21 |
| Total | 33 | 60 | 93 |
33 open-source tools are published in our index, all of them free (10) or freemium (23). The metered half of that column is empty: 0 open-source AI agents in our index bill by subscription or by usage; all 33 are free (10) or freemium (23). In the other direction, all 10 free tools are open source, and no closed-source tool in the index charges nothing at all.
Half of that is close to a tautology and half of it is genuinely informative. A tool whose source you can run yourself struggles to enforce a meter, so the empty cells in the open-source column are what you would expect. The informative half is on the closed-source side: 22 closed-source freemium tools are the only group in the index giving something away without being obliged to, and 10 of the 22 are coding agents. Research agents are the next-largest cluster at 5.
The 10 free tools have no paid tier of any kind on this reading: Aider, Juggler, Zot, AgentsKit, Goose, Microsoft AutoGen, the OpenAI Agents SDK, Pydantic AI, OpenClaw and GPT Researcher. That describes each vendor’s published pricing on 26 August 2026 and predicts nothing. Several tools now in the freemium column started here, and this exact list is unchanged from the edition we ran on 18 August, even though the classified census has grown by four listings since.
What would falsify the open-source finding is one open-source vendor putting a metered or flat-rate price on the hosted product with no free path, and that is a real possibility rather than a rhetorical one. This is a count of what 93 vendors published on 26 August 2026, not a law about how open source can be sold.
A free tier is not quite the same list as free-or-freemium
A separate Free Tier facet marks whether a listing offers any way to start without paying. 59 of the 93 classified listings offer a free tier of some kind, while 55 are classified free or freemium, so the two questions have different answers. Every free or freemium listing carries the facet on this reading, and the four extra are usage-metered tools with a self-serve entry plan: ElevenLabs Conversational AI, Manus, TinyFish and Webhound. Their headline enterprise meter is still a unit of work, so they sit in the usage-based column above. The mirror case the previous edition named, a genuinely free tool not marked with the facet, has none left here. The two lists differ in composition as well as in total, which is why we classify by the pricing table itself rather than by that facet.
Method
We read the pricing tiers on every published listing in this index on 26 August 2026, against each vendor’s own live pricing page, and classified each by its headline meter: free, freemium, flat subscription, or usage-based. The reading is pinned to one immutable collector run, 200142, so every figure above can be re-derived rather than re-counted: 94 listings published that day, 93 of them carrying a pricing model, 33 of those open source. Open-source status was determined for the same 93.
Since the edition we published on 18 August, which counted 89 listings, Clay left the index, with its usage-based enrichment pricing no longer checkable against a live page, and six listings joined: Kilo Code, MindStudio, Sim and Vellum as freemium, VoiceAgent as a subscription, and Agentix Labs, the agency listing this census excludes. That is 89 − 1 + 5 = 93 classified listings, out of 94 published. Five tools changed bucket in the same window: Auto-Respond from usage-based to subscription, Claude Code from subscription to freemium, Devin and Genspark from usage-based to freemium, and Vecbase from subscription to usage-based.
The field records the headline meter: the unit the bill scales with on the vendor’s primary paid path. Where a vendor publishes several tiers, the classification follows what those tiers are denominated in, not the cheapest entry point. A tool billed on a flat platform fee with metered credits layered on top, such as Kilo Code, is classified by the platform tier, not the credits. Vecbase was this page’s other example of that rule in August; both of its published tiers now price as “Usage-based” against a monthly credit allowance, so on this reading it is counted where its own tiers put it.
Usage-based does not mean expensive, and subscription does not mean predictable. The field describes the shape of the meter and says nothing about the size of the bill. A per-resolution price can be cheaper than a seat for a small team and far more expensive at volume, which is the point of the distinction rather than an objection to it. Our pricing FAQ works several of these tools through to actual figures, and the entry-price census counts who publishes a number at all.
What this does not say: these counts are a census of this index, not a survey of the market. They describe the 93 tools we classified as of 26 August 2026, out of the 94 published that day, and the date is part of the claim. The index grows, the buckets move under it: five did in the eight days before this reading. A figure here is a statement about run 200142 rather than about this minute. Corrections are welcome through our methodology page.
Get the next report
New agents and articles, by email, at most once a week. One-click unsubscribe.