⭐ Get the Agent OS + join 3,400+ founders inside the AI Profit Boardroom → Join AIPB ($69/mo)

GoldieBench Blog · 9 min read

Hark Pro Review: My Score, What I Tested, And The Models It Gets Compared With

Hark Pro review: I tested booking, Panels, inbox and Prep my day. It scores 8/10, plus pricing, privacy terms and how its rivals' models bench.

Hark Pro Review: My Score, What I Tested, And The Models It Gets Compared With — illustrated hero

My Hark Pro review score is 8 out of 10, because it is the easiest personal AI agent I have tested and it really does tasks on its own cloud computer.

Hark Pro is a personal agent from Hark Labs, the company run by Brett Adcock, who also founded Figure.

It reads your email, remembers your preferences and uses its own computer to book, plan and pay, and every feature is free.

It loses two points because it acts on real accounts, it does not talk back, and Hark has not named the model behind it.

I signed up on the free plan in launch week and tested it on camera, so this review is built on what I watched it do.

I'll give you the verdict, the scorecard, the test log, the pros and cons, the plan value, the terms and who should use it, and then the model data at the end.

The Hark Pro review verdict

QuestionMy answer
What is it?It is a personal AI agent that handles email, calendar, bookings and errands on its own cloud computer.
Who makes it?Hark Labs makes it, and Brett Adcock runs the company.
Where does it run?It runs on the web and in the Hark Pro apps for iOS and Android.
What does it cost?It is free, with Pro² at $20 a month and Pro³ at $100 a month for more usage.
My scoreI give it 8 out of 10.
Best forIt is best for people who want a personal agent working today without setup.
Skip it ifYou should skip it if you want to choose your own model.

If you want the setup steps first, read my Hark AI agent breakdown or the Hark Pro explainer.

This page is the review, so it focuses on how good it actually is.

How Hark Pro works

Hark Pro works in two halves.

The chat assistant works out what you mean and asks follow-up questions.

Handoff, which Hark calls its computer use agent, then does the clicking on a secure computer with a full browser that is individual to you.

Hark's site says Handoff can do research, place orders and build websites or slides.

Hark also says it asks for your approval before actions that require it.

That split matters for this review, because the brain and the hands are judged differently.

My Hark Pro review test log

I tested six things on the free plan.

TestWhat I didWhat happenedGrade
OnboardingI signed up and connected Gmail first, as Hark recommends.It showed my details and asked what it could take off my plate this week.A
Restaurant bookingI asked for an Italian restaurant next Wednesday.It asked for numbers and time, understood my typo'd "6pm, two people" and started booking until I said stop.A-
PanelsI asked for a productivity panel.It asked whether I meant meeting hours versus focus time or a to-do list, and offered ClickUp or Asana.A
Proactive inboxI let it read my email.It proposed action items that I could accept with a yes or a no.B+
Prep my dayI asked it to prep my day.It built an itinerary from my calendar and inbox.A
VoiceI spoke a request.It took voice input but replied only in text.C

The restaurant test is the one that tells you what kind of product this is.

When I connected Gmail, Hark Pro greeted me with, "Hi Julian. What's one thing on your plate this week I could take off?"

I typed, "Book an Italian restaurant for Wednesday next week."

It asked how many people and what time, and it understood my typo without any fuss.

Then it opened its own computer screen and started booking one of my favourite restaurants.

I told it to stop before it completed, because I didn't want a real booking for a demo.

Connected apps also worked directly inside Hark's own computer, which was easier than Grok Bot, where I often had to log in again.

The Prep my day example I saw said, "Your inbox is clean.

Do you want me to pull in the sales numbers before your team sales meeting at 4 p.m.?"

That is exactly the proactive behaviour you want from a personal agent.

The rest of the Hark Pro features I reviewed

FeatureWhat it doesMy score out of 10
HomeIt shows spaces, Panels and Action Buttons, and you can even pay bills from it.9
PanelsIt builds mini apps from a sentence, and the examples I saw tracked calendar events, Shopify packages and new Apple Podcasts episodes.9
ProjectsIt keeps long-horizon work, such as planning a big event, in threads that store files.8
IntegrationsIt connected Gmail, Slack, GitHub, LinkedIn, Notion, Amazon and PayPal in my test.8
WalletIt lets you add a wallet for payments.6
Side prompts and "show"It offers prompts like "quiet this" and pulls items into the chat with "show".7
Scheduled tasksIt lists every recurring job you set up in chat.8
Trip planningIt finds camping spots, writes a packing list, says when to leave and books flights.8
Attachments and voiceIt takes files and voice input but has no voice output.5

Hark's site also says Panels can connect to services such as Strava, Spotify and Venmo, so fitness, workout and calorie tracking panels are easy next steps.

Hark Pro review scorecard

CriterionScore out of 10
Ease of setup9
Gets real jobs done8
Proactiveness9
Control and safety7
Value for money9
Transparency5
Design and polish10
Overall8

I rounded down, because an agent that can touch your money has to earn trust over time.

Hark Pro pros and cons

These are the strongest points I found.

- The onboarding is the easiest I've had on any agent.

- It understood a messy request and acted on its own computer.

- Panels, the proactive inbox and Prep my day are genuinely useful every day.

- Every feature is included on the free plan.

These are the weakest points I found.

- It does not talk back, because voice is input only.

- Hark has not named the model behind its main assistant, and you can't swap it.

- It can act on real accounts, so a loose approval can cost you.

- It is brand new, so it has no long track record yet.

Is Hark Pro worth paying for?

Hark's pricing page says every plan has every feature, and upgrading only raises usage.

PlanPriceUsageMy value verdict
Free$0Base usage with no adsIt is the right start for everyone.
Pro²$20 a month2x more usageIt is worth it once you hit the free cap every week.
Pro³$100 a month10x more usageIt only makes sense for very heavy personal use.

Before launch, Brett Adcock posted that the first 100,000 sign-ups would get a paid plan free, though reports don't agree on the tier or for how long.

So stay on the free plan for two weeks and check your account before you pay.

The privacy and terms part of my Hark Pro review

I read the terms and privacy policy before connecting anything important.

- The terms say you must be at least 18.

- The terms say a transaction is authorised once you confirm it.

- The terms say you can give Hark standing permission for a category of transactions, which then doesn't need fresh approval.

- The terms say Hark isn't the merchant of record for most purchases, so refunds go through the shop or your card issuer.

- The terms cap Hark's liability for US users at the greater of what you paid in the last 12 months or $100.

- The privacy policy says your account won't be used to train AI models, and the terms say Hark's AI providers can't train on your content.

- The privacy policy says Hark hasn't sold or shared personal data in the previous 12 months.

My rule is simple: keep confirmations on, grant no standing permissions at first, and add your wallet last.

Who should use Hark Pro

You should use it if you have never had a personal agent working, because it is the easiest one I've tried.

You should use it if email and calendar eat your mornings, because Prep my day fixes that.

You should skip it for client work if you need to choose the model or keep data in your own stack.

A viewer asked whether, after Hermes, Muse, Dots, Grok, OpenBot and now Hark, the differences are big enough to set everything up from scratch again.

My answer is no for a working setup, so test Hark Pro on one job and only move if it clearly wins.

In my words from the video, it is a very simple, easy-to-use version of something like OpenClaw, straight out of the box.

Hark has also said it is working on its own hardware, and launch coverage points to 2027 for that.

The model question: what Hark has and hasn't said

This is the GoldieBench part of the review, and it comes last because most Hark Pro users never need it.

Hark Pro doesn't ship a model you can pick, and Hark hasn't named the model behind the main assistant.

What Hark has published is a claim about Handoff, its computer use agent.

Hark says Handoff scored 97.7 on Online-Mind2Web, ahead of GPT-5.4 at 92.8 and Claude Opus 4.8 at 84.1.

Those are Hark's own numbers from its own setup, so I treat them as vendor claims.

VentureBeat also reported that on WebTailBench v2, Handoff scored 68.6 against 72.3 for GPT 5.5.

So Handoff is strong at browsing, but it doesn't win every test.

How the models in Hark's comparison score on GoldieBench

GoldieBench gives every model the same one-shot prompt to build a game, a page, a simulation or a visual, then scores the result from 0 to 10.

Hark Pro itself is an app, so it isn't on the board, and neither is Handoff.

Of the models Hark compared Handoff against, only one is on the board.

ModelLink to Hark ProGoldieBench avgTasks scored
Opus 4.8Hark compared Handoff against it on browsing7.5147
GPT-5.4Hark compared Handoff against it on browsingNot on the boardNone
GPT 5.5VentureBeat reported it beat Handoff on WebTailBench v2Not on the boardNone

I won't guess scores for models that aren't on the board.

For context, the highest single models on the board right now are Claude Opus 5 at 8.27 and GPT-5.6 Sol at 8.16.

So the model Hark chose to beat on browsing is a solid builder on our board, but not a top-two model.

That doesn't contradict Hark's claim, because browsing a website and building one are different skills.

How Hark Pro's rivals' brains compare

Viewers keep asking how Hark Pro stacks up against the other personal agents, so here are their known brains on the board.

AgentBrainGoldieBench avg
Hark ProNot named by HarkNot on the board
Grok BotxAI's Grok models7.15 for Grok 4.7 and 8.09 for subscription Grok
Meta MuseMuse Spark7.55 for Muse Spark 1.2
OpenClawWhatever model you plug inIt depends on your choice

The scores for the known brains are close, which is the real lesson.

The differences you feel between personal agents come mostly from setup, design and how each one handles your accounts.

That's why my Hark Pro score rests on my tests, not on a leaderboard.

If you want a model you control completely, the local models board shows which free models are good enough to run on your own machine.

What GoldieBench can and can't tell you about Hark Pro

GoldieBench measures how well a model builds working things in one shot.

It doesn't measure booking a table, reading an inbox or clicking through a checkout.

In my restaurant test, understanding the request was the assistant's job and doing the booking was Handoff's job.

Neither half is something the board tests directly, so the fairest test is still the one I ran: give it a real job and watch.

Hark Pro isn't the only agent with its own cloud computer, and my Manus video below shows another take on the same idea.

Also On Our Network

Test it on one real job, keep your approvals tight, and you'll have the evidence that matters most in any Hark Pro review.

FAQ

Is Hark Pro worth it?

Yes on the free plan, because every feature is included. I score it 8 out of 10, with points lost for no voice output, an undisclosed main model and approvals you must manage carefully.

How much does Hark Pro cost?

Hark's pricing page lists a free plan with every feature, Pro² at $20 a month for 2x more usage and Pro³ at $100 a month for 10x more usage.

Is Hark Pro on GoldieBench?

No. Hark Pro is an app and Handoff is a computer use agent, while GoldieBench scores models on one-shot builds. Hark hasn't named the model behind its main assistant, so there is no score to show for it.

How does Hark's Handoff compare with Opus 4.8?

Hark says Handoff scored 97.7 on Online-Mind2Web against 84.1 for Claude Opus 4.8. On GoldieBench, which tests building rather than browsing, Opus 4.8 averages 7.51 across 47 tasks.

Is Hark Pro safe to connect to my wallet?

Hark says it asks for approval before actions that need it, but its terms say a confirmed transaction is authorised and allow standing permissions. Keep confirmations on and add your wallet last.

The same stack Julian uses

Run this stack yourself.

Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 3,400+ founders shipping with it every day all live inside the AI Profit Boardroom.

3,400+founders
258documented wins
38countries
$69/momonthly