Every brand publishes a suction figure and almost none of them publishes what it means. Here is what each number is measuring, and what to look at instead.
Disclosure — some links on this page go to Amazon, and we earn a commission if you buy through them. It costs you nothing extra and it does not change what we recommend or in what order. The full disclosure.
Published suction on current robot vacuums runs from 5,300 Pa to 22,000 Pa, and every one of those is a peak figure measured at the nozzle with nothing in the way. It tells you roughly where a model sits in its maker's own range. It does not let you compare brands, and it says nothing about carpet.
Carpet is where the suction arms race happens, so it is worth being clear about what the numbers are. A pascal rating is a pressure measurement taken under ideal conditions in the manufacturer's own lab, with the manufacturer's own method. There is no shared standard for it. When Roborock says 22,000 Pa and eufy says 8,000 Pa, you are not looking at a comparison — you are looking at two companies measuring different things and printing both on a box.
That does not make the figure useless. Within one brand's range it is a reliable ranking. Across brands it is a rough tier indicator and nothing more. What follows is every figure we could find published, with the wording each maker used.
Stated suction across every model we cover
22 of 23 models publish this figure. Every number below links to the page it came from.
Compiled from the manufacturers’ own published material on 2026-09-01. We have not measured any of these figures ourselves — see how we pick. Found an error? Tell us and we will correct it.
Two entries in that table are worth pausing on. iRobot publishes no absolute figure at all for the Roomba Combo j9+ — its suction is quoted only as being 100 percent stronger than an older iRobot model, which is a comparison you cannot carry across to any other brand. Shark does the same thing by omission on both of its models. Whatever you think of those machines, they cannot be placed on this axis, and we are not going to invent a number so the table looks tidy.
What actually decides carpet performance
If the suction figure is only a tier indicator, the things that genuinely separate a good carpet robot from a bad one are these, roughly in order:
Carpet detection and boost. The robot has to notice it has driven onto carpet and increase power. Every model here claims this; what varies is how it detects the transition and how much headroom it has left when it does.
Brush contact with the pile. A brush that sits above deep pile is not cleaning it. This is why chassis-lift models matter on high pile: they change the ride height, not just the climb ability.
Mop management. On a mop combo, the mop must lift clear or detach before the robot crosses carpet, or you are damp-mopping a rug. Published lift heights run from 10mm to 17mm, and some models publish nothing.
Whether it can get onto the carpet at all. A thick rug on a hard floor is a threshold. If the edge is higher than the robot's published climb, the rug never gets cleaned and the robot spends the run bumping into it.
High pile is a different category
There is a real break point somewhere around 20mm of pile depth. Below it, most current robots manage. Above it — deep shag, thick wool, a heavy sisal — the robot starts riding on top of the fibers instead of pressing into them, the wheels lose traction, and side brushes flick debris rather than sweeping it. No suction figure fixes a robot that is floating.
This is the one case where chassis-lift hardware is doing something other than climbing thresholds. Roborock's AdaptiLift and Dreame's ProLeap both raise the body, which changes how the brush meets deep pile. That is covered in detail on high-pile carpet.
Area rugs, fringe and the edge problem
An area rug on a hard floor creates three separate problems: a lip the robot has to climb, a fringe that behaves exactly like a mop pad to a rotating brush, and a boundary the mop has to know about. Fringe in particular is the single most common way people permanently damage a rug with a robot vacuum, and it is entirely avoidable with a no-go zone.
Ranked on the published specifications below. Tap a row to jump to the full write-up.
#
Product
Best for
Suction
Price
1
RoborockNo photo
Roborock Qrevo CurvXThe highest published suction figure here paired with the highest published mop lift, which is the combination a mixed-floor home actually needs.
Dreame X50 UltraThe only model here that legs itself over a step rather than driving at it, and the only one that publishes the condition attached to its own headline number.
ECOVACS DEEBOT X8 PRO OMNIA rotating roller mop that washes itself continuously while it cleans, instead of pushing one increasingly dirty pad around the house.
Dreame L40s UltraStates 40mm of obstacle clearing and a fully detangling brush, which covers the two problems that actually send robots back to the shop.
Roborock Qrevo CurvPublishes both a standard-threshold and a double-layer threshold figure, which is the number most sill-blocked buyers are actually shopping for.
The only model here that legs itself over a step rather than driving at it, and the only one that publishes the condition attached to its own headline number.
60mm over a series of steps, 42mm over a single stepDreame's own condition, quoted: "The gap between steps should be greater than 4cm. For a single obstacle crossing and a single vertical step, the highest obstacle crossing height is 4.2cm." The 6cm headline therefore does not describe a single door sill.Source: Dreame X50 Ultra product page — ProLeap step climbing (retrieved 2026-09-01)
Publishes the caveat as well as the headline, which is rarer than it should be
20,000 Pa stated peak suction and a 395ml on-board bin
Stated 220 minutes of runtime in quiet mode
What it does not
The legs add height and mechanical complexity to a machine that lives under furniture
10.5mm mop raise is modest next to the 17mm figures from Roborock
Who it is for
Split-level rooms, a sunken living room, or a doorway sill that has beaten every robot you have owned.
Skip it if
You read the 6cm headline and stopped there. Dreame's own footnote puts a single vertical step at 42mm, and 6cm only applies to a series of steps with more than 4cm between them.
40mm (1.6in) double-layer, 30mm (1.2in) standardRoborock's own wording is that it can overcome standard thresholds up to 3cm and navigate double-layer thresholds up to 4cm. Stated maxima assume a clean square edge; a rounded or worn sill behaves differently.Source: Roborock Qrevo Curv product page — threshold crossing (retrieved 2026-09-01)
5,300 Pa is the lowest stated suction figure anywhere in this comparison, and it is an older generation. On hard floors with a thin rug that is fine and it is often cheap. On anything you would describe as carpet, you are buying the wrong end of the only spec that matters here.
Skip this one
Shark PowerDetect 2-in-1
Not because it is a bad machine — the LiDAR and the pad-washing base are real. But Shark publishes no suction figure for it at all, and this is a page about comparing suction figures. If you are shopping specifically for carpet performance, buy from a maker that will tell you a number you can hold them to.
The short version
Buy within a brand on its suction figure, not across brands. Confirm the robot can climb onto every carpeted surface in your house before you compare anything else. If your pile is deep, prioritize a model that lifts its chassis over one with a bigger pascal number. And if the model also mops, check the published lift height in the mop lift comparison — a wet rug is a worse outcome than a slightly dusty one.
Go deeper on carpet
Start with the suction comparison if you are choosing between models, or with the high-pile page if your carpet is thick enough that the robot rides on top of it.
Beaching on the edge and pulling at fringe are different problems. Both are fixable for nothing.
Questions people actually ask
+ Do robot vacuums work on carpet?
Yes, and daily running is what makes them effective — a robot that runs every day on medium pile keeps ahead of debris in a way a weekly upright cannot. Deep pile is the exception: above roughly 20mm the robot starts riding on the fibers and results fall off sharply.
+ What suction do I need for carpet?
There is no honest answer in pascals, because the figure is not standardized between brands. Within one brand's range, more is better on carpet. Across brands, use it as a tier indicator and decide on carpet boost, brush contact and climb height instead.
+ Do robot vacuums go over rugs?
They climb onto them if the edge is below the robot's climb height, which for most current models is somewhere between 20mm and 40mm where it is published at all. A thick rug with a stiff bound edge is effectively a threshold, and a rug that slides is worse than one that is too tall.
+ Will a robot vacuum ruin a rug fringe?
It can. A rotating brush treats fringe exactly like it treats hair, and a long fringe can be pulled hard enough to damage the weave. The fix is a no-go zone drawn along the fringed edge, which every mapping robot here supports. Full detail on rugs and fringe.
+ Is carpet boost worth it?
It is table stakes rather than a differentiator — every mapped model on this page claims some form of it. What matters is whether the robot detects the transition reliably in your house, which depends on the carpet as much as the robot.
+ What is the best robot vacuum for high pile carpet?
One that raises its chassis rather than one with the largest suction figure. Riding height decides whether the brush is in the pile or above it, and no amount of airflow compensates for a brush that is not touching the fibers. See high-pile carpet.
