AI Agent Pricing Index 2026: 16 Platforms, Checked Monthly

Every comparison post in this category quotes a price. Most quote it once, on the day it was written, and never look again. A quarter later the figure is wrong and the post still ranks.

Three of my own posts did exactly that. They quoted prices that had moved within a quarter of publishing, and I only found out when a reader emailed me. The monthly re-read below exists so the correction happens on a schedule instead of by embarrassment.

This page is the repair I wanted for my own writing. One row per tool: starting paid price, free tier, what the meter counts, what the first paid plan includes, and a link to that tool's dated price history. As of 11 September 2026 it covers 16 AI agent and automation platforms.

Every figure was read on the vendor's own pricing page, on that date, at the URL listed in the sources at the bottom. No estimates. No averages.

Behind the table sits a longer record: 314 archived snapshots since January 2024, grouped into 97 distinct price periods. That is how a row can say when a price last moved.

Copy it. That is the point.

The index is licensed CC BY 4.0, so any writer can lift the rows into their own comparison and link back here. I would rather see my numbers in a competitor's post than watch another invented figure spread. For context around them, cheapest AI agent platforms and the glossary sit one click away.

The index

Prices read on each vendor's pricing page on 11 September 2026; annual-billing prices where the page defaults to them. A "page figure" tag means the figure is on the page but the plan mapping is my read; "not published" means the page shows no price. "Last price change seen" is the month the plan-price set changed in the monthly Wayback Machine captures, with how many captures exist and where the archive stops; it links to the dated history with archived sources. Where a vendor prices in euros the figure is kept in euros. Per-seat prices are marked. Tools whose page shows no figure are listed as not published rather than guessed.

Two labels in that table carry a precise meaning. A row reads not published when the vendor's page showed no figure at all on the date it was read, because the plan sits behind a sign-in or a contact form. A row reads first paid tier on the page when the figure is printed but the vendor does not tie it to one named plan, so the mapping is the reader's to make, not mine. Both labels exist so a gap in the record looks like one.

Each tool's row links to its dated history page in the last column; Make and Gravity have too few captures for one yet.

How to read the units

The number on a pricing card tells you less than the unit underneath it. What the plan counts decides the invoice. A $9 plan and a $16 plan are 78% apart on the card and the wrong way round on a four-step flow, because 1,500 runs costs 6,000 Make credits against 1,500 Activepieces credits. Nobody publishes that conversion, so every "cheapest" list in this category is comparing labels.

Three families sit in this index, and one workload separates them: 1,500 runs a month of a flow with four billable action steps.

Zapier and Make count every action

Zapier charges one task per step, and more for AI steps. Make charges one credit per module action. Under both, four billable action steps across 1,500 runs costs 6,000 units a month. Add two steps and the same 1,500 runs cost 9,000. Zapier does not bill the trigger, filter or path steps, so a four-step Zap with a trigger bills three tasks a run (4,500 a month), while Make bills the trigger module too, so its 6,000 stands.

Zapier's Professional plan starts at $19.99 a month on annual billing, on a slider beginning at 750 tasks, so 6,000 tasks sits well above the floor. Make's $9 plan carries 5,000 credits, so the same workload pushes you up the credit slider before you start. Make is the thinnest row in the index: two archived captures exist, January 2024 and February 2025, and they match. That is absence of evidence, not evidence of stability. Zapier pricing explained unpacks what a task really is.

Per run or execution: n8n, Activepieces, Pipedream

One unit per whole run here, whatever the step count. The same workload costs 1,500 n8n executions, 1,500 Activepieces credits and 1,500 Pipedream workflow executions. Four steps or fourteen, the number is identical.

n8n Starter is €20 a month billed annually and includes 2,500 executions, so the workload fits the entry plan. Activepieces Plus is $16 a month billed yearly for 10,000 credits, then $0.007 a credit. Pipedream's first paid tier on the page is $29 a month, and its credits track compute time, so a slow step costs more than a fast one at the same run count.

What does a seat or a flat plan actually buy?

It buys a person's access, with the usage pooled behind it. Lindy is $29.99 per seat a month for 3,000 credits, and one task can draw 2 credits or 1,000 depending on the work. Run count barely predicts that bill. Task weight does.

Gumloop is $37 a month for 20,000 credits drawn down per node run, with an 8% orchestration fee on top when you bring your own model keys. Work that percentage out before you move anything high-volume across.

1,500 runs of a four-step flow is 6,000 node runs against Gumloop's 20,000 credits on Pro. On Lindy the same runs at 2 to 250 credits each land anywhere from 3,000 to 375,000 credits against a 3,000-credit Plus seat, so the credit-per-task range is the number to check before comparing Lindy on price.

Gravity is the flat end of the family: $20 a month on Autopilot worldwide, INR 1,999 a month in India, with a lot of usage included. India also has Minipilot at INR 399 a month with a smaller allowance. The first agent is free, no card, and you buy more usage if you run out. My product, so read the row with that bias attached: 1,500 runs of a four-step job does not move the invoice.

The same job priced five ways, across per execution, per step, per run and per seat, is in 9 best n8n alternatives. When a vendor here shuts down the same arithmetic decides where its customers go, as the Relay.app alternatives post worked through in September 2026.

Change log

A change-log entry is one dated, sourced sentence about a single move: what changed on which vendor's page, in which month, with a link to the archived snapshot behind it. No commentary on why.

1. 14 September 2026: Relay.app, closed for paying customers (free accounts ended 15 August 2026); no longer tracked (source)
2. 1 July 2026: Zapier, the page's plan-price set moved from $13/$33/$66 style annual figures to $13.33/$19.99/$33.33 style figures (monthly-equivalent display), per our July snapshot (source)
3. 1 October 2025: Zapier, annual-billing figures ($13, $33, $66 a month) appeared next to the monthly ones on the page, per our October 2025 snapshot (source)
4. 1 July 2025: Activepieces, the page's plan set changed to $0, $25 and $150 a month, per our July 2025 snapshot; by September 2026 Plus is $16 a month yearly with a per-credit overage (source)
5. 4 April 2025: n8n, a €60 figure joined the page's plan set alongside €20, €24 and €50, per our April 2025 snapshot (source)

The weekly watch finds these by fetching each tracked pricing page, extracting the plan figures on it, and comparing that set with the last one on file. A different set raises a flag. Nothing is logged until somebody opens both snapshots side by side, because a discount ribbon and a price cut look identical to a script.

Method

Here is exactly what produces the table, so you can decide how far to trust it.

The current row comes off the vendor's own pricing page. Every figure was read on 11 September 2026 at the URL listed in the sources. Where a page defaults to annual billing, that is the figure recorded, and the row says so. Euro prices stay in euros, with no conversion applied.

The history comes from monthly Wayback Machine snapshots. Up to one capture per tool per month, wherever the archive has one. That comes to 314 snapshots across the 16 tools, unevenly spread: 32 for Taskade back to January 2024, 2 for Make, and nothing in the last eight months for Make, Bardeen or Voiceflow.

A script does the extraction, not the judgement. It pulls the plan figures printed on each archived page, groups consecutive snapshots that share a figure set into one period, and dates the boundary where the set changes. That produced 97 price periods across the 16 tools.

A person reads every flagged change before it enters the index. That is me, most months. Extractors happily pick up banner offers and unrelated numbers, so an automated diff is a prompt to go and look.

Two things this method refuses to do. It does not guess a price when the page hides one behind a sign-in or a sales form, which is why four rows carry a label where a number would go. And it does not publish average pricing, because the average of four entry plans is a figure nobody is charged.

One honest limit, stated plainly: I read the pages, and nothing here is a test of the products. This index tells you what each vendor charges and what they count. Whether the thing works is a different question.

Use this data

The rows are free to reuse, including commercially, including by companies listed in the table.

If you maintain a comparison post of your own, the history link is the durable thing to borrow. Prices here move two or three times a year. A dated link survives that; a hardcoded $19.99 does not.

Corrections are welcome, and early ones are best. If a vendor's page shows something different from a row here, the fix lands in the next monthly refresh with a change-log line. Corrections: support@gravity.fast, with the vendor page and the date you read it. Gravity's own pricing page is held to the same standard as the other 15.

FAQ

Sources