Hark Pro Claimed "Best on Earth." Then People Tested It.
Launch day: "the best AI on Earth at using websites." Day two: people put it to work.
Hark’s claims (Hark’s own numbers):
- 97.7 on a web-use benchmark
- About 10x cheaper than the top models
What early testers found:
- One reviewer gave it 6 real jobs (dinner, flowers, comparison shopping, flights, a memory test, a meter audit). Website navigation held up impressively.
- But it confidently stated a baggage policy that wasn’t true.
- Gizmodo: a simple DoorDash order took longer than doing it yourself.
- Another early user hit browser windows they couldn’t click into.
- And one early tester called it the best interface in agents.
Verdict: split. Great at clicking, not at knowing. Use it for busywork and double-check any fact before you act on it. Early sign-ups get Pro free until Nov 6.
Would you let an AI book your next flight?
Gizmodo launch-day hands-on (search “Gizmodo Hark Pro”)
Sources & links 4
More on AI Agent News, AI Safety & Regulation News