Hark, an AI startup founded by Brett Adcock, has unveiled Handoff, a computer use agent designed to perform complex tasks across the open web without human intervention. The system can independently order food via DoorDash, book flights on United and Delta, and message job candidates on LinkedIn, using dedicated virtual computers with full browser access, file systems, and terminals. Handoff achieved a score of 97.7 on the Online-Mind2Web (OM2W) benchmark, surpassing OpenAI's GPT 5.4 at 92.8, Anthropic's Claude Opus 4.8 at 84.1, and Google's Gemini 2.5 Pro at 69. The model is priced at $0.18 per million input tokens and $2.37 per million output tokens, significantly below GPT 5.5's $5 and $30 rates. It also delivers responses in 0.8 seconds per turn. Users can link existing accounts so Handoff accesses saved addresses, payment methods, and browsing history. Sign-ups are now open at hark.com, with general availability expected later this month as part of Hark's initial software rollout.
Handoff claims superior performance and lower costs, but its real-world reliability will depend on how well it navigates unscripted website changes or login failures. The absence of public third-party verification for its benchmark score and pricing means early adopters may need to test claims firsthand before trusting it with sensitive tasks. If the efficiency holds, it could shift competition among AI agents toward affordability and speed.
Editorial note: AI-assisted opinion, not established fact. Full disclaimer →
