Hark previews Handoff browser agent after topping Online-Mind2Web

Why it matters

Handoff turns Brett Adcock's personal-AI plan into a testable software product before Hark ships hardware, but its verified advantage is a narrow 0.4-point benchmark lead.

Brett Adcock (@adcock_brett) introduced Hark Handoff on August 5th, putting his second major AI venture into research preview with a browser agent designed to order food, book travel, shop and complete other tasks across the open web.

Handoff is the first concrete product from Hark, the personal-AI lab Adcock started while continuing to run humanoid robot maker Figure. Adcock, who grew up on a farm in central Illinois, previously built recruiting marketplace Vettery and co-founded electric-aircraft developer Archer Aviation before moving into humanoid robotics. Hark extends the same full-stack approach into consumer AI: models, software and eventually purpose-built devices.

Hark says Handoff creates a dedicated virtual computer for each request, including a browser, file system and terminal. The agent controls a cursor and keyboard rather than depending on integrations with individual services, allowing it to work across websites that lack consumer-facing APIs. Hark's demonstrations include placing DoorDash orders, shopping at Target, reserving restaurants through OpenTable, messaging candidates on LinkedIn and booking flights.

Users can connect existing accounts, giving Handoff access to saved addresses, preferences and purchase histories. That design makes reliability and account security central product requirements. A browser agent that can merely find a flight is materially different from one trusted to log in, select an itinerary and make a purchase.

A verified lead, with limits

Handoff scored 97.7% on the human-evaluated Online-Mind2Web benchmark, according to the official leaderboard. The test covers 300 tasks across 136 live websites. Handoff recorded 100% on tasks classified as easy, 95% on medium tasks and 100% on hard tasks in an evaluation dated August 4th.

That result narrowly passed Yutori's Navigator n1.5, which scored 97.3% in June, and an ACT-2 system powered by GPT-5.4, which scored 92.7%. Careerflow.ai conducted the human evaluations for all three entries.

The leaderboard supports Hark's claim that Handoff currently holds the top verified Online-Mind2Web score. It does not establish Handoff as the best internet-use model across every test or production setting. Hark's broader "best ever" description stretches beyond the scope of the independent result.

Hark also published results for WebTailBench v2 and an internal evaluation. Handoff scored 68.6 on WebTailBench, behind GPT-5.5 at 72.3 and ahead of Claude Opus 4.8 at 66.9. On Hark's internal test, Handoff led with 83.2, compared with 80.5 for Opus 4.8 and 75.2 for GPT-5.5. Hark ran those comparisons through its own evaluation harness and used an internal large-language-model judge, leaving Online-Mind2Web as the independently evaluated portion of the release.

Hark's software entry point

Adcock launched Hark publicly in March with $100 million of his own capital, according to reporting at the time. Hark has described its larger goal as a personal intelligence system with speech, vision, persistent memory and custom hardware. Adcock told Bloomberg that Hark was developing a family of devices for individuals and homes, separate from phones, wearables and smart glasses.

Handoff gives that plan an immediate software entry point. It places Hark inside a browser, where users already manage travel, shopping, communication and other daily work, before Hark attempts to shift those interactions onto its own devices.

Hark said the results came from six months of work on computer-use agents. The current model reflects post-training through supervised fine-tuning and reinforcement learning. Hark has since moved into mid-training and plans to begin pre-training later in 2026, a sequence that lets its researchers develop data pipelines, evaluations and infrastructure before training a model from the ground up.

Handoff remains a research preview rather than a generally available service. Hark is accepting applications for beta access and says its platform will become available by the end of summer. The release gives Adcock a verified benchmark lead and a product surface for his personal-AI thesis. The production test will begin when outside users hand Hark their accounts, purchases and real-world tasks.