Relevance AI vs Gumloop vs Lindy pricing: what going over your plan costs on each

Short answer: the three cheapest self-serve agent platforms give three completely different answers to the same question. Go over your plan on Relevance AI and the next 1,000 Actions cost $80, which is more than ten times what an included Action costs on Pro. Go over on Gumloop and nothing happens at all until you switch overage on, after which credits bill at $0.005, about 2.7 times the included rate, under a cap that defaults to $5,000. Go over on Lindy and there is no overage to buy: it pauses. Every figure below was read from the vendors' own pages between September 5 and September 8, 2026, from a United States address.

Last updated September 2026. We sell a flat rate AI agent with no credits and no Action meter, so we have an interest in how this comes out. Check every number against the vendor pages we name, and check the date on anything else you read.

The number that decides your ninth month is not on the pricing page

Every comparison of these three products argues about the entry price. Relevance AI is $19 a month, Gumloop is $37, Lindy is $29.99 per user. Those numbers are correct and they will tell you almost nothing about what you end up paying, because all three sell you an allowance and all three behave differently the moment you exhaust it. That behaviour is what separates a $37 plan from a $5,037 one, and it is the part buried in documentation rather than printed on a pricing page.

Here is what each vendor actually does at the edge of its plan, from its own published rates.

Read down the last four rows rather than the first two. Those are the rows that describe what happens on a month you did not plan for, and they are the rows that no headline comparison includes.

Relevance AI: the steepest cliff, softened by the best rollover policy

Relevance AI runs two meters at once. Actions count tool runs, one per run, and Vendor Credits cover the model and tool compute underneath. The two behave nothing alike.

Vendor Credits are the most buyer friendly thing in this comparison. They cost $20 per 10,000, which is $0.002 each, and that is exactly the rate at which your plan includes them: Pro's 10,000 a month are described by Relevance AI as $20 of value, Team's 35,000 as $70. Buying more costs precisely what being given them was worth. They also roll over indefinitely while your subscription stays active, and on any paid plan you can bypass them entirely by supplying your own model keys. No other self-serve vendor here matches that.

Actions are the opposite. A top-up is $80 per 1,000, or $0.08 an Action, sold only in thousands and only on a paid plan. Pro includes 2,500 in a $19 plan. Attribute every cent of that $19 to Actions and pretend the $20 of Vendor Credits it also hands you is worth nothing, which deliberately overstates the included price, and an included Action still works out at $0.0076. So the first Action past your allowance costs more than ten times the one before it, with no change in what you are doing.

There is a specific trap in that arithmetic and it is worth writing down. On monthly billing, Pro at $29 plus four Action top-ups comes to exactly $349.00 and gives you 6,500 Actions. Team, billed monthly, is $349.00 and gives you 7,000 Actions, 25,000 more Vendor Credits, three more Build Users, 45 End Users where Pro has none, calling and meeting agents, an analytics dashboard and priority support. Identical money. On annual billing the crossover comes a step earlier, at the third top-up. We work through the whole ladder on our Relevance AI pricing page, including why the vendor's own pricing page no longer carries a single dollar figure while its documentation publishes all four tiers.

One more line to price in: Relevance AI states that if a tool fails, it still counts as one Action. On a clean workload that is noise. On anything touching a rate limited API, a scraped source or a legacy CRM, a 10% failure rate is a 10% surcharge on the meter that has the margin in it.

Gumloop: nothing happens by default, which is the safest failure mode of the two metered options

Gumloop sells one credit at $0.005 and includes 20,000 of them in a Pro plan that starts at $37. Its own documentation is unusually candid about what that means: the 20,000 are described as 7,400 credits at the $0.005 list price plus 12,600 bonus credits. Multiply 7,400 by $0.005 and you get exactly $37.00, so 63% of the allowance is a promotional bonus. An included credit therefore costs $0.00185 and an overage credit costs $0.005, a ratio of 2.7.

That is a real cliff, but it is a quarter as steep as Relevance AI's and it does not arrive uninvited. Gumloop's overage is off until you enable it on the Usage and Limits page, and once enabled it is always capped on Pro, with a default ceiling of 1,000,000 overage credits per period, which is $5,000, adjustable downward. Agents stop at the cap. Several guides ranking for Gumloop's pricing tell readers the opposite, that Pro defaults to unlimited overage, and they are wrong in the direction that costs money.

Two Gumloop details are worth stealing for your evaluation of any platform. Failed tool calls are not charged at all, which is the direct opposite of Relevance AI's rule. And there is an 8% orchestration fee applied to reasoning, tool calls and compute on every agent chat, which is the only percentage fee on top of a meter we have found across nine platforms. Our Gumloop pricing page prices a month of work in the order Gumloop actually bills it.

Lindy: no overage exists, and that is a feature

Lindy sells credits per seat and spends them from one shared workspace pool: $29.99 for 3,000 credits on Plus, $99.99 for 15,000 on Pro, $199.99 for 35,000 on Max. When the pool empties, Lindy pauses credit-using actions. There is no overage rate, no top-up to buy and no invoice you did not authorise. Of the three, it is the only one where the worst case is your agents stopping rather than your bill moving.

The cost of that safety is forecasting. Lindy publishes bands rather than rates: an everyday ask is 2 to 250 credits, deep work is 250 to 1,000, a big build is 1,000 to 2,500. A Plus seat's 3,000 credits is therefore either 1,500 everyday asks or twelve of them, and no plan page can tell you which. Credits do not roll over, and a seat is created by anyone who mentions Lindy in a Slack channel. The detail on all of it is on our Lindy pricing page.

The same month of work, priced in each vendor's own units

Take a modest but real workload: 2,000 agent runs a month, each making three tool calls, each taking about a minute of processing, with an 8% failure rate and roughly $40 of model compute. Here is what that costs on each, using published rates only.

Relevance AI. Three tool runs times 2,000 runs is 6,000 Actions, plus 8% failures is 6,480. Pro cannot absorb that: 2,500 included plus four top-ups is $339 a year-billed month for 6,500 Actions. Team at $234 includes 7,000 and covers it outright, and its 35,000 Vendor Credits cover the $40 of compute with room to spare. $234 a month.

Gumloop. A chat with three tool calls (3 credits), a minute of compute (5 credits) and moderate reasoning (say 12 credits) is 20 credits, plus the 8% orchestration fee, rounded up to 22. Times 2,000 chats is 44,000 credits against an allowance of 20,000, so 24,000 credits of overage at $0.005 is $120 on top of the $37 plan. $157 a month, and $0 if you never enable overage, because the agents simply stop at 20,000.

Lindy. This one genuinely cannot be priced from published information. At the bottom of Lindy's everyday band, 2 credits a run, 2,000 runs is 4,000 credits, which is two Plus seats at $59.98. At a still modest 20 credits a run it is 40,000 credits, which needs more than one Max seat at $199.99. Somewhere between $60 and $400, and the vendor cannot narrow it for you either.

The honest caveat is the interesting part: an Action, a Gumloop credit and a Lindy credit measure different things and do not convert. Relevance AI charges per tool run regardless of model cost. Gumloop charges for the model at its own token price and adds a percentage. Lindy charges an opaque number per ask. Anyone presenting these three as a single price comparison is quietly inventing a conversion rate that does not exist, which is why the table above compares behaviour instead.

Which of the three we would pick, and when we would pick none of them

If your workload is spiky and you want protection from your own mistakes, Lindy's pause is the safest design here and the seat price is the whole price. If your workload is high volume and tool heavy, Gumloop is the cheapest per unit of the three, the failed calls are free, and the cap is real. If you want to run your own models at wholesale and keep credits that never expire, Relevance AI's Vendor Credit policy is genuinely the best in the category, provided you buy the right plan first rather than topping up into it.

All three share one cost that never appears in a pricing table: they are builders. You design the agent, connect its tools, test it, and fix it when an integration changes underneath you. For a technical operator that is the point. For a small team without one it is a real line item, whether it shows up as your own evenings or as hiring someone to build and maintain it. Price that in before you compare plan tiers, because it is usually larger than the difference between them.

The alternative is to stop buying units altogether for the part of the work that never varies. Recurring, rule shaped jobs run the same way every week, and paying a meter for them means paying again every week for an outcome whose shape you already know. That is the half we built WorkAgent for, at $149 a month flat: no Actions, no credits, no seats, no orchestration fee, and no charge for a run that failed. A heavy month and a quiet month cost the same. Judgement work, one off investigations and genuinely novel builds are where a metered builder earns its money, and we would not pretend otherwise. Most teams need some of both, and the useful exercise is deciding which half is which before you sign anything.