Online-Mind2Web: 300-task live-web agent benchmark on Kernel#43

Open

rgarcia wants to merge 9 commits into

hypeship/cua-bench-online-mind2webfrom

hypeship/bench-online-mind2web

Commits on Jun 28, 2026

Add online-mind2web Harbor adapter (Kernel env + cua agent, WebJudge)

rgarcia
and
claude
committed
online-mind2web: opus judge + live-smoke fixes

rgarcia
and
claude
committed
online-mind2web: carry dataset level into [metadata].difficulty

rgarcia
and
claude
committed
online-mind2web: default WebJudge to OpenAI o4-mini

rgarcia
and
claude
committed
online-mind2web: back WebJudge model with pi-ai

rgarcia
and
claude
committed
online-mind2web: send o-series reasoning effort, drop temperature in the pi-ai judge

rgarcia
and
claude
committed
online-mind2web: assert openai:o4-mini default in adapter test

rgarcia
and
claude
committed
online-mind2web: harden judge reward write and threshold parse

rgarcia
and
claude
committed
online-mind2web: correct the judge-backbone README to the pi-ai transport

rgarcia
and
claude
committed