DEV Community: Mobilebox

Best cloud phones for AI agents (2026)

Mobilebox — Sat, 14 Nov 2026 17:56:17 +0000

Disclosure: we build Mobilebox, one of the services in this post. Written with AI assistance from our own published data; every number carries its date and source, and corrections go to hello@mobilebox.dev.

The best cloud phone for an AI agent depends on the job. Limrun gives an agent a working Android screen fastest, about 20 times faster to a first screenshot than Mobilebox in our test (0.98 s against 19.4 s), and adds iOS simulators and remote Xcode. Phone Harness and Mobilerun rent real iPhones; Mobilerun also gives agents phone numbers and SMS. Clousd lists the lowest price per minute. Mobilebox fits when every task or pull request gets its own phone that is started, checked, screenshotted and stopped for a few cents.

A cloud phone for an agent is a phone in a data center that your agent drives over an API or MCP; see what a cloud phone for AI agents is. We make Mobilebox, so it is one row below, judged like the rest. Facts come from each vendor's own pages, reviewed on November 10, 2026; speed numbers are ours, from the November 8 benchmark.

The shortlist

Service	Good for	The phone	Price, as published	A stopped phone keeps
Limrun	Coding agents that build and check iOS and Android apps	"remote Xcode, iOS & Android Simulators"	Android $0.04 a minute on Entry ($20 a month)	Not stated; instances end when idle
Phone Harness	A real iPhone for your agent, or your own phone	Cloud emulators; real cloud iPhones, $80 a month	5¢ a minute, boot free	"apps and logins"
Mobilerun	Phone numbers, SMS and residential IPs	"on-demand Android phones"; physical iPhone add-on	$20 for 600 min; then $0.03 a minute	"apps and data"
phonebox.dev	Signed-in Android phones, API key only	"Cloud Android devices"	$0.06 a minute; Starter $20 a month	"apps, files and signed-in accounts"
Mobilebox	A phone per task or PR, from CI or a cloud agent	Cloud Android 14; iPhone and iPad $0.05 a minute	$0.01 a ready minute, per second, boot free	Apps, data and logins, free
Clousd	The cheapest minute, phones in many countries	"a full Android phone in the cloud"	$0.004 a minute, at most $0.80 a day	"apps and data"
Ramus	Coding agents checking their own APK	"Cloud Android emulators for coding agents"	30 min a month free; $2 and $10 plans	Nothing: "A fresh device every session"

Measured: start, tap and a 10-minute session

Three of these let us measure on a free tier, with the same steps and the same APK, 10 sessions each, from one client in Nuremberg, Germany.

Service	Start to first screenshot	Tap to screenshot	Install a 4 MB APK	10 minutes
Limrun	0.98 s	0.38 s	0.85 s	$0.40
Mobilebox	19.4 s	1.45 s	1.17 s	$0.10
Phone Harness	23.0 s	1.91 s	3.57 s over ADB	$0.50

Medians; the benchmark has p90s, the CSV and the scripts. Limrun was fastest on every speed metric, very likely because its phones are booted before you ask. Mobilebox was the cheapest for 10 minutes and the only one where we measured park and resume: 11.5 seconds back to a screenshot, the installed app still there in 10 of 10 runs. For a one-minute task a 20-second start is a third of the wall clock; for a 30-minute session it hardly matters.

We didn't measure Mobilerun (sign-up is Google, GitHub or Discord only), phonebox.dev (the free week needs a card), Clousd or Ramus. Their rows are as published.

Device farms: real devices for test suites

Device farms rent real phones to run test suites across many models. They are built for QA teams more than for an agent that acts step by step.

Farm	Price, as published	Agent access
BrowserStack	App Live from $39 a month, App Automate from $199 a month, billed yearly	Remote MCP server with OAuth
AWS Device Farm	$0.17 a device-minute, or $250 a month per unmetered slot	Its API; no Device Farm MCP server found
Firebase Test Lab	$5 an hour per physical device, $1 per emulated one	Test matrices; Google says it shuts down on September 30, 2027

BrowserStack's terms forbid publishing performance results, so we didn't time it. If your app checks for real hardware, use a physical phone; see real devices or cloud phones.

Or run the phone yourself

A local emulator with mobile-mcp costs nothing on a laptop with hardware acceleration and a screen. In CI it is slower: on a GitHub-hosted runner the emulator took 50 seconds to a first screenshot with KVM and 13.5 minutes without. See cloud phone or local emulator.

Pick by situation

The agent can't wait for a boot: Limrun.
A real iPhone: Phone Harness or Mobilerun by the month; Mobilebox iPhones by the minute.
Numbers, SMS and residential IPs for the agent: Mobilerun.
A fresh phone for every pull request, logins kept when you want them: Mobilebox. Budget about 20 seconds per start and about 10 cents per 10 minutes.
The cheapest minute: Clousd.
A clean device every time, no state: Ramus.

How we picked

A service is on the list if an agent can create and drive a phone in the cloud through it today, over MCP or an API, with public prices. Each row quotes the vendor's own page as of November 10, 2026; prices are list prices that day, as published. Speed numbers are only the ones we measured. If a row is wrong or out of date, write to hello@mobilebox.dev and we will fix it. Six of these side by side on sign-in and endpoints: cloud Android phones for AI agents, compared. The MCP side, including local servers: the best Android MCP servers.

Originally published at mobilebox.dev/blog/best-cloud-phones-for-ai-agents.

How fast do cloud Android phones for AI agents start? (Nov 2026 benchmark)

Mobilebox — Sat, 14 Nov 2026 17:56:15 +0000

On November 8, 2026, Limrun gave our script a usable cloud Android phone in under a second (0.98 s) at the median, Mobilebox in 19.4 seconds and Phone Harness in 23 seconds. Mobilebox, which runs cloud Android phones for AI agents at $0.01 a minute, was the cheapest for a 10-minute session ($0.10) and the only one where we measured park and resume; Limrun was also fastest from tap to screenshot (0.38 s, against 1.45 s for Mobilebox and 1.9 s for Phone Harness).

These are our own measurements: 10 sessions per service from one client in Nuremberg, Germany, on each vendor's free tier or trial, with the same steps and the same APK. The benchmark page has every metric, the p90s, the CSV and the scripts.

The numbers

Service	Cold start to first screenshot	Tap to screenshot	Install a 4 MB APK	10-minute session
Limrun	0.98 s (p90 1.07)	0.38 s (p90 0.40)	0.85 s	$0.40 on Entry
Mobilebox	19.4 s (p90 19.5)	1.45 s (p90 1.49)	1.2 s	$0.10
Phone Harness	23.0 s (p90 70.0)	1.9 s (p90 2.4)	3.6 s, uploaded over ADB	$0.50

Medians, p90 by nearest rank, nothing removed. Prices are list prices on the day: Limrun $0.04 a minute on its Entry plan (pricing), Mobilebox $0.01 (pricing), Phone Harness $0.05 (limits and credits).

Why the starts differ so much

Limrun's create call returned a ready phone in about a third of a second, so its phones were very likely booted before we asked. Mobilebox boots a phone per request, and Phone Harness's timings look the same: Mobilebox in a narrow 18.9 to 19.5 seconds across all 10 runs, Phone Harness in two clusters, about 21 seconds and about 67 seconds. Phone Harness's own docs say "about 80 seconds", so its faster runs beat its published number.

For an agent the difference is real. A one-minute task spends a third of its wall clock waiting on a 20-second start, and almost none on a one-second start. For a 30-minute session the start barely matters, and the price per minute and what a stopped phone keeps matter more.

Tap latency includes the network

We timed one tap plus one full screenshot through each vendor's own interface: Limrun's SDK, the Mobilebox REST API, and ADB for Phone Harness, which has no tap endpoint. Network distance counts twice. Our client was 8 ms from Mobilebox's API, about 50 ms from Limrun's and about 170 ms from Phone Harness's servers at US West addresses. Phone Harness's 252 ms taps are mostly that round trip; Mobilebox's 517 ms taps, from 8 ms away, are not.

What a stopped phone keeps

Mobilebox parked in about 1 second and came back to a screenshot in 11.5 seconds, with the installed APK still there in 10 of 10 runs. Phone Harness documents a saved profile phone that "comes back the same way next time"; we didn't test it. Limrun's Android instances end when they are deleted or time out, so there is nothing to resume.

What we left out

BrowserStack, Genymotion and LambdaTest forbid publishing performance results or competitor use in their terms, so we didn't measure them. Mobilerun, AWS Device Farm, Firebase Test Lab and phonebox.dev need an OAuth or Google sign-in or a payment card, which this run didn't use. The benchmark page lists each reason and their published prices.

Run it yourself

The scripts are in fifteenA/mobilebox/benchmarks, one per service. Each prints one line per measurement, appends it to a CSV and deletes what it created. Results will differ from your location and on another day; that is why the client, date and raw samples are published.

The same test on a CI runner (from the benchmark page)

The benchmark page also ran a local emulator on a GitHub-hosted runner (ubuntu-latest, API 34, 5 sessions, November 8, 2026):

Setup	Start to first screenshot	Tap to screenshot	Install the APK
Local emulator, KVM	50.0 s (p90 50.3)	1.03 s	6.09 s over ADB
Local emulator, no KVM	13.5 min (n=1; the 2nd boot didn't finish in 15 min)	18.7 s	2.1 min over ADB
Mobilebox, called from the same runner	20.3 s (p90 20.6)	2.09 s	1.17 s

With KVM the emulator also needed about 75 seconds per job to download the image and create the AVD before it booted.

Originally published at mobilebox.dev/blog/cloud-android-start-time-benchmark.