Gemini completed all 10 tasks in a reported hands-on comparison of four Android assistants on a Pixel 9 Pro XL running Android 17. Perplexity completed 8; ChatGPT and Claude completed 5 each. The results cover one phone, one attempt per task and the specific paid plans and models used in the test—not Android assistants across every device or setup.
What the test asked each assistant to do
Each assistant received the same ten spoken requests, word for word, with one attempt per task. The chores covered phone controls, messaging, a calendar event, music, screen reading, directions, weather and travel advice: set a 12-minute timer and a 6:45 a.m. alarm, send a text saying the user was running 10 minutes late, add a Friday lunch to the calendar, play Phoebe Bridgers on Spotify, turn on the flashlight, summarize an article already open on screen, find a gas station, check Miami’s weekend forecast, and answer a question about currency and tipping in Cozumel.
The test used ChatGPT Plus with GPT-5.6 Sol (Android), Claude Pro with Claude Opus 5.5, Gemini AI Pro with Gemini 3.8 Flash, and Perplexity Pro with its automatic “Best” model selection. Tasks were graded as done, assisted, handed off or failed; “done” meant the assistant completed the request without extra taps.
How the four assistants compared
| Assistant | Tasks completed (of 10) | Weather response began |
| Gemini | 10/10 | About 2 seconds |
| Perplexity | 8/10 | About 45 seconds |
| ChatGPT | 5/10 | About 10 seconds |
| Claude | 5/10 | About 3 seconds |
The weather times are approximate response-start observations from this test, not standardized latency measurements. Completion totals make the clearest distinction: Gemini handled every listed request, while the other assistants’ results depended more on the kind of task.
Screen reading and directions
Gemini was the only assistant in the test to summarize the article already open on screen. ChatGPT, Claude and Perplexity did not complete that request: each asked for the article’s text or a screenshot instead.
For directions to a gas station, Gemini completed the task and Perplexity opened Google Maps with a route loaded. Claude handed the request off by suggesting map apps. ChatGPT failed after repeated prompts for precise location.
Phone controls, messages and music
Gemini and Perplexity completed the flashlight request; ChatGPT and Claude handed it off. All four completed the calendar task. For the text message, Gemini sent it after asking for confirmation. Perplexity asked for the phone number after it failed to find the contact, while ChatGPT and Claude handed the task off.
ChatGPT set its timer and alarm inside its own app rather than Android’s Clock app; the timer later appeared as a notification. Claude used Android’s Clock app, and Gemini and Perplexity completed the timer-and-alarm request. For Spotify playback, Gemini completed the task, Perplexity was marked done, Claude provided an artist-page link that needed a tap to start playback, and ChatGPT handed the request off.
Weather response and lock-screen use
The Miami weather response began in about 2 seconds for Gemini, 3 seconds for Claude, 10 seconds for ChatGPT and 45 seconds for Perplexity. Those timings come from a single reported comparison, so they offer context for the interaction rather than a repeatable speed ranking.
Lock-screen behavior also differed. ChatGPT worked from the lock screen. Gemini worked after a tap to confirm the request, while Perplexity asked the user to unlock the phone; Claude did not work from the lock screen.
Permissions and choosing a default
In the reported setup, all four assistants had access to notifications, location, the calendar and contacts. Gemini also needed screen access for some actions. Making an app the default assistant did not grant those permissions by itself.
On the tested Pixel, the selection path was Settings > Apps > Default apps > Digital assistant app. For a reader who mainly wants on-screen summaries and routine controls, Gemini handled the broadest range of these particular tasks. Perplexity completed more chores than ChatGPT or Claude in the test, but its weather response took longer to begin and it required a microphone tap before each question. Claude completed the same number of tasks as ChatGPT, with a different balance of phone control and interaction.
