Research reportai-agents
AI Agent Entry Prices: 40 of 92 Publish One, and the Median Is $20
We read the pricing table of all 92 tools in our index on 19 August 2026. 40 publish an exact flat monthly price in dollars, and the median is $20 a month.
Of the 92 AI agent tools in The Agents Index, 40 publish an exact flat monthly plan price in US dollars, and among those 40 the median cheapest paid plan is $20 a month. That is the answer to the question a buyer with a budget line actually asks, which is not what pricing model a vendor uses but what the smallest cheque looks like.
We have published a census of the pricing models too, and it answers a different question. This edition counts the numbers on the pricing tables themselves, re-derived on 19 August 2026 against every published listing. Fewer than half the index answers the entry-price question at all. Among the ones that do, the answers pile up on a single figure.
The headline count
Of the 92 tools in our index, 40 publish an exact flat monthly plan price in US dollars. Among those 40, the median cheapest paid plan is $20 a month.
The denominator matters more than the number, so it is worth being blunt about it. This is a statement about the 40 tools that publish a comparable monthly price. It is not a statement about the index, and it is certainly not a statement about the market. The other 52 are accounted for below, and they are not missing data. Most of them are vendors that genuinely do not sell this way.
Within those 40:
| Cheapest paid monthly plan | |
|---|---|
| Lowest | $5 |
| Median | $20 |
| Highest | $2,417 |
Ten of the 40 are priced at exactly $20 a month: Browserbase, Cursor, Factory AI, Julius AI, Letta, MindStudio, Perplexity, Vecbase, Warp and Windsurf. That is the largest cluster in the distribution by a wide margin, and it is why the median sits exactly on $20 instead of somewhere near it. Both middle values of the 40 are $20, so the median does not depend on which of them you take. Twenty-six of the 40 are at or under $25. Eight are at or over $100.
Why only 40, and what the other 52 do instead
An absence here is not a zero. The tools outside the 40 are outside it for reasons that are themselves the finding. This accounts for all 92 listings:
| How the vendor prices its cheapest paid plan | Tools |
|---|---|
| Exact flat monthly price | 41 |
| Qualified monthly price (“from $X”, “~$X”, “$X starting”) | 5 |
| Per-seat or per-agent monthly price | 8 |
| Usage-metered rates only | 14 |
| Annual or quarterly contract price only | 3 |
| Priced, but not as a subscription | 1 |
| Free only, no paid tier carries a published price | 11 |
| Custom or contact-sales only | 8 |
| No pricing table recorded in our index | 1 |
| Total | 92 |
The top row is 41 rather than 40 because one of those tools, n8n, prices in euros (€20/month Cloud Starter). Folding a euro price into a dollar median would need a conversion we have not done and could not date, so n8n is counted as an exact monthly price and excluded from the dollar figures.
Five tools publish a monthly figure that is a floor rather than a price: Devin and Manus both lead with “from $20”, LiveKit Agents publishes “$50/month starting”, Rox publishes “$100/month starting”, and Antigravity CLI publishes “~$19.99”. Those are real published numbers in the same range as the rest, but they are not exact, so they sit in their own row.
One listing in our index has no pricing table at all
The last row of that table is a gap in our own research, and it is worth naming rather than rounding away. Goose, the open-source coding agent from the Agentic AI Foundation, was published on 17 August with an empty pricing-tier table. Our structured facets record it as free, but this census is derived from the pricing tables and nothing else, so Goose is excluded from every figure here rather than assumed into the free bucket.
That is the difference between a tool we know charges nothing and a tool whose pricing we have not written down. One of those is a finding and the other is a to-do list item. Goose is the second, and it is the only one of the 92.
The two ends of the table are open source and enterprise sales
The 11 free-only tools are open-source projects with nothing commercial priced above the free tier: AgentsKit, Aider, CrewAI, GPT Researcher, Juggler, Microsoft Agent Framework, Microsoft AutoGen, the OpenAI Agents SDK, OpenHands, Pydantic AI and Zot. Several of them do sell something above the free tier, but none of those tiers carries a published figure, which is why they land here and not in the priced rows.
The 8 custom-only tools are the opposite end, enterprise platforms where every path to a working product runs through a sales conversation: Ada, Artisan, Decagon, Glean, Moveworks, PolyAI, Qualified and Sierra. The single tool priced outside any subscription is Webhound, which sells from $1 with no recurring plan at all.
The category ladder is the real finding
Splitting the 40 by category produces a steep and fairly clean ordering. The category a tool sits in predicts its entry price better than anything else we hold, and the ordering has not moved since the last edition even though the corpus has.
| Category | Tools | Lowest | Median | Highest |
|---|---|---|---|---|
| Coding agents | 15 | $8 | $19.00 | $100 |
| Agent platforms | 6 | $20 | $39.50 | $149 |
| Research & data agents | 6 | $16 | $22.50 | $500 |
| Agent frameworks | 4 | $20 | $100.00 | $250 |
| Voice agents | 3 | $5 | $6.00 | $19.99 |
| Sales & marketing agents | 3 | $167 | $600.00 | $2,417 |
| Support agents | 2 | $10 | $79.50 | $149 |
| Agent tools | 2 | $20 | $29.50 | $39 |
(Botsify belongs to two categories and is counted once in each, so the rows sum to 41 against 40 distinct tools. The same convention applies in our deployment census.)
Read this table with the tool counts in view. Only coding agents have enough priced tools for a median to describe a category rather than a handful of vendors. At two, three or four tools the median is arithmetic on a very small sample, and the range is the more honest column.
Coding agents have converged on the price of a chat subscription
With that caveat stated, the coding-agent row is still the one worth citing. Fourteen of the 15 coding agents that publish a flat monthly price fall between $8 and $25. Codex CLI is the floor at $8, then Cline at $9.99, then GitHub Copilot, Junie, OpenCode and Zed at $10, then Claude Code at $17, then Kilo Code at $19, then the $20 wall, then Bolt.new and Replit Agent at $25.
The fifteenth is Augment Code at $100 a month, four times the next-highest coding agent and the only tool in the category priced above the band. A category with 25 tools in it, competing hard on capability, has otherwise converged on the price of a consumer chat subscription.
The sales and marketing row is the mirror image, and its tools are the same ones we found were cloud-only with no self-hosting path at any price: Clay at $167, Amplemarket at $600, and 11x at $2,417/month for its Julian Growth chat plan, the highest published monthly figure anywhere in the index. Nothing in that category is priced like a tool you expense.
What moved since the 17 August edition
The 17 August edition read 38 of 87, with the same $20 median. Five listings have been published since, which moves the denominator to 92. Two of the five publish an exact flat monthly price and join the count: MindStudio at $20 and Kilo Code at $19. Of the other three, Sim prices per user, TinyFish meters by request, and Goose carries no pricing table.
One vendor changed its own pricing table in the meantime. Rox raised its Individual tier from “$50/month starting” to “$100/month starting”, which leaves it in the qualified row where it already sat but doubles the figure. The $20 cluster grew from nine tools to ten, and the count of priced coding agents from 14 to 15, with the $8-to-$25 band still holding for every one of them except Augment Code.
What this does and does not tell you
If you are being quoted $1,200 a month for a coding agent, the 15 priced tools in that category say that is well outside what anyone publishes. If you are being quoted $1,200 a month for a sales agent, the three priced tools there say nothing of the kind. That figure is inside the observed range, and the published prices in that category span a factor of nearly fifteen.
What an entry price cannot tell you is total cost. Several of the flat monthly plans meter something on top. Botpress is $89/month plus AI spend, Auto-Respond is $10/month plus $1 per minute of talk time, and the credit-pooled plans bill overages at a premium. An entry price is the cost of the door, not the cost of the room.
Method
Every figure comes from the pricing_tiers table on each listing, which records the vendor’s own published pricing page, re-verified on the schedule described in our methodology. Clay’s $167 Launch tier, for example, is read from Clay’s pricing page. Counted 19 August 2026 against all 92 published listings.
The entry price for a tool is the lowest exact flat monthly fee, in US dollars, that the vendor publishes for the product. Where a tool publishes no such fee, it is classified by the cheapest monthly figure it does publish, and where it publishes no monthly figure at all, by whatever it publishes instead. One further rule decides the borderline cases: a tool is excluded from the exact-price count when its own cheapest plan is priced as a floor, because the exact figure further up the ladder is then not the price of entry. That rule is what puts Devin in the qualified row despite its $500 Team plan carrying an exact monthly price.
Four exclusions are applied, each of which removed real rows:
- Free tiers are excluded. A $0 plan is a free tier, and we count those separately. The entry price is the cheapest plan you pay for.
- Per-seat prices are excluded, because a $20/seat price and a $20 platform price are not the same number, and averaging them merges two axes. Eight tools price only per seat or per agent and appear in that row of the table above.
- Non-monthly units are excluded: per minute, per token, per hour, per resolution, per GB, per credit, per conversation, and annual or quarterly contract prices. One trap worth naming is Cresta, which publishes $150,000 on a “/year (12-month contract)” basis. A naive filter for the word “month” reads that as a monthly price. It is not one, and it is not in these figures.
- Promotional first-month rates are excluded in favour of the standing rate. OpenCode’s Go plan is $5 for the first month and $10/month after, and it is counted at $10.
Three further rules decide the cases that a reader is most likely to disagree with, and each of them moves at least one tool:
- “$X starting” is read the same way as “from $X”. Both are floors, which is why LiveKit Agents and Rox sit in the qualified row rather than the exact one.
- A flat fee for the product counts as a flat price even when it bundles seats. Agno at $150/month with four seats included is a platform price, not a per-seat price. The same reading admits Amplemarket and Augment Code.
- A tool’s entry price is its cheapest paid plan, even when that plan meters something on top. Auto-Respond is counted at $10/month rather than at its $19 second tier, and Kilo Code at the $19 Kilo Pass Starter rather than at its per-user Teams tier.
Medians of an even number of values are the midpoint of the two middle values, which is why the research row reads $22.50 against underlying prices of $20 and $24.99.
Where a vendor publishes both a month-to-month rate and a lower annual-commitment rate, we use the figure the vendor’s own pricing table leads with, which is inconsistent between vendors because their pages are. This does not move the headline. Substituting every annual-commitment rate for its month-to-month equivalent leaves the median at $20, and the two middle values of the distribution are both $20 either way.
Prices change, and this is a snapshot with a date on it. For tools whose pricing model changed during 2026, see 9 AI agent tools that changed how they charge. For a closer look at five individual vendors’ bills, see our pricing FAQ. For how many publish a complete price list rather than just an entry point, see our price-transparency census.