Hark, a newly established artificial intelligence venture led by entrepreneur and roboticist Brett Adcock, has officially announced the launch of its primary product, Handoff. According to VentureBeat, Handoff is designed as a computer use agent (CUA) engineered to autonomously handle complex tasks across the open internet, such as booking travel, ordering food, and managing professional communications.

The startup claims the agent is among the highest-performing tools currently available for web-based autonomy. Hark reports that Handoff achieved a score of 97.7 on the Online-Mind2Web (OM2W) benchmark. This figure is compared against scores of 92.8 for OpenAI’s GPT 5.4, 84.1 for Anthropic’s Claude Opus 4.8, and 69 for Google’s Gemini 2.5 Pro.

Model Performance and Pricing Comparison

Beyond performance metrics, the company emphasizes the cost-efficiency of the agent, citing per-turn latency of 0.8 seconds. Each interaction with Handoff involves the deployment of a dedicated virtual computer environment, including its own file system, terminal, and web browser. Users may integrate existing accounts, allowing the agent to utilize saved payment methods and address data. Hark's research indicates that while 75% of daily screen time is spent within a browser, fewer than 1 in 1000 websites provide public APIs, creating a distinct niche for agents capable of navigating standard web interfaces.

Public registration for the platform is currently open via the company website, with wider availability scheduled for later this month. Despite the performance claims, industry observers note that the company’s benchmark comparisons rely on previous-generation models, excluding current leaders like GPT-5.6 and Claude Opus 5, as well as open-source variants such as DeepSeek V4.

Why It Matters

The introduction of Handoff signals a shift toward practical, agentic AI that prioritizes browser-based interaction over proprietary integrations. By functioning as a surrogate user on existing websites, Hark bypasses the bottleneck of limited API availability that currently restricts most automation software. If Handoff maintains its performance benchmarks at the stated price points, it could significantly lower the barrier to entry for individual and enterprise-level automation, forcing incumbent AI providers to optimize for operational cost and browser-based efficiency rather than solely relying on conversational performance.