{"benchmark":"Browser Use Pi engine path vs ego (lite) browser automation","date":"2026-09-11","environment":{"hardwareArchitecture":"arm64","operatingSystem":"macOS 26.5.1 (25F80)","egoVersion":"0.5.1.2","egoChromiumVersion":"152.0.7977.54","agentBrowserVersion":"0.37.1","targetSite":"https://the-internet.herokuapp.com"},"methodology":{"runsPerCase":3,"timerScope":"Mechanical browser execution after the target page loaded. Includes a browser observation, actions, waits, and final verification. Excludes initial navigation and model deliberation.","piPath":"agent-browser 0.37.1 direct CLI, the browser engine wrapped by pi-agent-browser-native. Pi, its selected model, and extension presentation overhead were not timed.","egoPath":"ego-browser operating one dedicated ego (lite) Space through semantic browser locators and snapshots.","successRule":"The final page state had to match the case-specific assertion. A dispatched action without a verified state change was not counted as success.","limits":["This is a browser-tool benchmark, not an end-to-end Pi agent benchmark.","The target is a deterministic public test site; results do not predict arbitrary production websites.","The benchmark does not cover CAPTCHA, MFA, checkout, file upload, or long-running agent planning.","The two tools have different process and selector models, so timings measure the tested operator stacks rather than raw Chromium speed."]},"summary":[{"case":"form-login","egoPasses":"3/3","egoMedianMs":887,"agentBrowserPasses":"3/3","agentBrowserMedianMs":1016},{"case":"dynamic-controls","egoPasses":"3/3","egoMedianMs":7412,"agentBrowserPasses":"3/3","agentBrowserMedianMs":6852},{"case":"add-remove-elements","egoPasses":"3/3","egoMedianMs":4244,"agentBrowserPasses":"3/3","agentBrowserMedianMs":1435},{"case":"table-max-due-extraction","egoPasses":"3/3","egoMedianMs":7,"agentBrowserPasses":"3/3","agentBrowserMedianMs":368},{"case":"session-survives-reload","egoPasses":"3/3","egoMedianMs":343,"agentBrowserPasses":"3/3","agentBrowserMedianMs":664}],"runs":{"ego":{"form-login":[941,859,887],"dynamic-controls":[7384,7412,7434],"add-remove-elements":[3362,4244,4250],"table-max-due-extraction":[7,6,7],"session-survives-reload":[429,341,343]},"agentBrowser":{"form-login":[1016,949,1064],"dynamic-controls":[6747,6950,6852],"add-remove-elements":[1435,1190,4356],"table-max-due-extraction":[244,368,2771],"session-survives-reload":[664,630,894]}},"excludedPilots":["Three initial agent-browser reload runs waited for a transient login-success flash that disappears after reload. They were discarded before timing because the wait condition did not match the published success rule; the final runs verified the secure URL and welcome state.","An initial ego add/remove run used an ambiguous selector after three identical Delete buttons appeared. The final runs used an explicit first-match selector and verified that exactly one button remained."]}