GoldieBench Blog · 9 min read
Hark Pro Explained: The Plans, The Setup, Every Feature, And The Model Reported Inside
Hark Pro is a free personal AI agent from Hark Labs. What Pro² ($20) and Pro³ ($100) add, how to set it up, every feature, and the model reported inside.

Hark Pro is a free personal AI agent from Hark Labs that connects to your email and apps, remembers your preferences and does real jobs on its own cloud computer.
Hark Labs was founded by Brett Adcock, who also founded Figure and Archer Aviation, and Hark announced a $700 million funding round in May 2026.
Hark Pro launched on 6 October 2026 on the web, iOS and Android.
Every feature is on the free plan, and the two paid plans, Hark Pro² and Hark Pro³, only buy you more usage.
I tested it on the free plan on camera, so in this post I'll explain what it is, what each plan gives you, how to set it up, every feature I tested and its limits.
After that, I'll cover the question builders keep asking me, which is what model sits inside it and how that model scores on GoldieBench.
What Hark Pro is
Hark describes Hark Pro as "the operating system for your life".
You chat with it like a messaging app, and it remembers what you tell it.
It reads your connected email and apps, so it can suggest the next job before you ask.
It then uses Handoff, its own cloud computer, to sign in to websites and complete tasks like purchases and forms.
Hark says each Handoff request gets a dedicated virtual computer with a browser, a file system and a terminal.
When I opened it, it felt like the Apple of AI agents, because the design is that clean.
Hark Labs, Inc. is based in San Jose, California, according to its terms, and those terms say you must be at least 18 to use it.
Hark Pro plans: free, Pro² and Pro³
| Plan | Price | Usage | Features |
|---|---|---|---|
| Hark Pro | Free | The base allowance | Every feature |
| Hark Pro² | $20 a month | 2x more usage | Every feature |
| Hark Pro³ | $100 a month | 10x more usage | Every feature |
Hark's pricing page says all features are offered in all plans and upgrading only increases usage.
It also gives two examples of scale, saying the $20 plan can get organic produce delivered 40 times and the $100 plan can plan 50 weddings.
Those are illustrations, not hard limits.
If you run out, Hark's terms say it may offer paid Top-Ups.
Automatic Top-Ups only apply if you switch them on, and Hark's guide recommends setting a spending ceiling when you do.
Unused Top-Ups are not refundable unless your local law gives you that right.
My advice is to run a normal week on the free plan, move to Pro² only if you hit the limit, and leave Pro³ for people whose whole day runs through Hark.
How to set up Hark Pro
Setup took me a few minutes.
First, sign up at hark.com or in the Hark Pro app with Google, Apple or an email address.
Second, connect your email, because Hark recommends doing that first, and I connected my Gmail.
Third, check the details Hark shows you, such as your location.
Fourth, answer its opening question, which for me was, "Hi Julian. What's one thing on your plate this week I could take off?"
Fifth, connect the apps you use, such as Slack, GitHub, LinkedIn, Notion, Amazon and PayPal, and add your wallet only when you're ready.
Hark also says it connects to Google and Outlook through their APIs, lists Google Workspace and Microsoft 365 as direct Services, and supports MCP for other tools.
How Hark Pro works: the restaurant test
I typed, "Book an Italian restaurant for Wednesday next week."
Hark asked how many people and what time, and it understood my typo'd reply of "6pm, two people".
Then it started booking one of my favourite restaurants on its own computer screen, and I told it to stop before it completed.
Connected apps work directly inside Hark's own computer, which is easier than Grok Bot, where I often had to log in again through its computer.
I called it a very simple, easy-to-use version of something like OpenClaw straight out of the box.
Every Hark Pro feature I tested
| Feature | What it does |
|---|---|
| Chat and memory | One persistent thread that learns your preferences, and you can ask it to forget specific memories |
| Handoff | Hark's cloud computer that signs in to websites and completes tasks |
| Home | Spaces, Panels and Action Buttons on one screen, and you can even pay bills from it |
| Panels | Mini apps you create by typing, such as a productivity panel with ClickUp or Asana, Shopify packages, calendar events or new Apple Podcasts episodes |
| Action Buttons | One-tap cards such as paying a bill or cancelling a subscription |
| Projects | Dedicated threads that store files for long-horizon work like planning an event |
| Proactive inbox | Reads your email and proposes action items for a yes or no |
| Side prompts and "show" | Prompts like "quiet this" and "close this notification", and a show function that pulls items into the chat |
| Attachments and voice | Files and voice input are accepted, but there is no voice output |
| Scheduled tasks | One section that manages every recurring job you set up in chat |
| Specialties | Multi-step workflows Hark memorises for jobs you repeat |
| Prep my day | Builds your itinerary from your calendar and inbox |
| Trip planning | Finds camping spots, writes a packing list, says when to leave and books flights |
When I asked for a productivity Panel, Hark asked whether I meant meeting hours versus focus time, or to-do tasks.
Fitness, workout and calorie-tracking Panels are easy next ideas, and Hark's guide says you can use "Reply" on a Panel to redesign it.
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.?"
Hark also says it's designing its own hardware and training smarter models.
Hark Pro limits and safety
Hark's terms say it will ask you to confirm before it spends your money unless you tell it otherwise.
They also let you give standing permission for a category of transactions, and every purchase in that category then counts as authorised by you.
Hark isn't the merchant of record, so refunds go through the shop or your card issuer.
Hark says it will never sell your data or share it with advertisers, and cards and passwords sit in the encrypted Secured by Hark vault.
Its privacy policy says it doesn't use your inputs, outputs, memory or connected account data to train AI models.
It doesn't speak back, because voice is input only.
It's also built for one person's life admin, so it isn't a team tool or a crew of agents.
How to get the most from your Hark Pro usage
Because every plan has the same features, the real skill is spending your usage well.
First, batch small requests into one clear message, because one precise job is cheaper than five vague ones.
Second, turn any job you repeat into a Specialty, so Hark doesn't have to relearn it every week.
Third, use scheduled tasks for routine checks, such as a weekly bill review, instead of asking by hand.
Fourth, keep Handoff for jobs that really need a browser, and use plain chat for quick questions.
Fifth, keep each big job in its own Project, so the files and context stay together and Hark doesn't waste effort rebuilding them.
Sixth, review your Home screen every week and delete Panels you no longer look at.
If you switch on auto top-up, set the spending ceiling before anything else.
Who Hark Pro is for
Hark Pro is for busy people who lose hours every week to small admin jobs.
It suits founders who want a morning brief, parents juggling bookings and bills, and anyone who travels often.
It's a good first agent for people who have never had one, because the onboarding is the easiest I've tried.
It's less suited to builders who want to pick their own model or run a pipeline of agents for client work.
For those people, Hark Pro works best as a personal lane next to the tools they already control.
And because it doesn't let you choose the model behind it, the next question is what that model actually is.
Which model is inside Hark Pro?
Hark Pro doesn't ship a model you can pick, and Hark hasn't officially named the model behind its main assistant.
Here is what is actually known.
Hark says Handoff is its own computer use model, which it post-trained itself.
Hark claims Handoff took the top spot on the Online-Mind2Web human evaluation leaderboard and beat GPT 5.4 by 8 points and Opus 4.8 by 2 points on average.
Those are Hark's own numbers, so treat them as vendor claims.
TestingCatalog reported that Hark's configuration labels Claude Opus 5.5 for the main assistant, task agents and website building.
The same report listed Qwen3.6-27B and Gemini 3.5 Flash Lite for browser logins and GPT-6 Astra for browser challenges.
TestingCatalog itself warned that these labels don't confirm which model handled any real request.
So the honest answer is that the main brain is undisclosed, and the Claude Opus 5.5 label is the strongest public clue.
How the models linked to Hark Pro 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 rendered result from 0 to 10.
Hark Pro is an app, so it isn't on the board, but the models linked to it are.
| Model | Link to Hark Pro | GoldieBench avg |
|---|---|---|
| Claude Opus 5.5 | Reported label for the main assistant | 7.57 |
| Opus 4.8 | Hark compared Handoff against it | 7.51 |
| Gemini 3.6 Flash | The nearest Gemini Flash on the board, since 3.5 Flash Lite isn't benched | 7.08 |
| Qwen 3.7 | The nearest Qwen generalist on the board, since Qwen3.6-27B isn't benched | 7.00 |
GPT-6 Astra and GPT 5.4 aren't on the board, so I won't guess their scores.
For context, the top single models on the board are Claude Opus 5 at 8.27 and GPT-5.6 Sol at 8.16.
So if the Opus 5.5 label is right, Hark Pro's main assistant sits in the solid upper-middle of the board rather than at the very top.
What the bench means for your Hark Pro plan
Here is the link between the model and your plan that most people miss.
The paid plans don't buy you a smarter model, because every plan has the same features.
They buy you more runs of the same brains.
So if Hark Pro gets a job wrong on the free plan, Pro² won't fix it, and you should rewrite the instruction instead.
GoldieBench also doesn't test browsing or bookings, so it can't score Handoff directly.
It tells you how strong the reasoning and building brains are, which matters for the websites and documents Hark builds.
If you want a brain you fully control, the local models board shows which free models are good enough to run on your own machine.
My launch-day breakdown is in the Hark AI agent post, and my scored verdict is in the Hark Pro review.
Should you switch agents for Hark Pro?
A viewer listed Hermes, Muse, Dots, Grok, OpenBot and now Hark, and asked whether the differences are big enough to set everything up again.
The bench says the brains behind these agents are close, so I wouldn't switch for the model alone.
Switch only if Hark Pro does a real job better than the agent you already use.
If you've never had a personal agent working for you, Hark Pro is the easiest place to start.
Also On Our Network
- Agent Operatorshow to set up Hark Pro step by step
- AI Income Deskseven business jobs Hark Pro can handle
- AI Tool VerdictHark Pro's three plans compared and scored
- agentos.guidea five-minute Hark Pro quick-start for your Agent OS
- agentos.guidethe Hark AI agent quick-start on agentos.guide
Start on the free plan, judge it on real jobs, and use the bench to understand the brains inside Hark Pro.
FAQ
What is Hark Pro?
Hark Pro is a personal AI agent from Hark Labs that launched on 6 October 2026. It connects to your email and apps, remembers your preferences and uses its own cloud computer, called Handoff, to book, buy, plan and pay for things on the web, iOS and Android.
How much do Hark Pro² and Pro³ cost?
Hark Pro is free, Hark Pro² costs $20 a month for 2x more usage, and Hark Pro³ costs $100 a month for 10x more usage. Hark says all features are offered in all plans.
Does Hark Pro have its own model?
Hark Pro doesn't let you pick a model, and Hark hasn't officially named its main assistant's model. Handoff, its browser agent, is Hark's own post-trained computer use model, and TestingCatalog reported a Claude Opus 5.5 label for the main assistant.
Is Hark Pro on GoldieBench?
No. Hark Pro is an app, not a model. Claude Opus 5.5, the label reported for its main assistant, scores 7.57 on GoldieBench.
Does a paid Hark Pro plan use a better model?
Hark says every plan has the same features and upgrading only increases usage, so a paid plan gives you more runs rather than a different model.


