Under development. These results may change as we optimize the app and test it in different lab conditions. Each row shows the phone it ran on and the date it was shared.
Loading shared results…
| Model | Score | Tokens/s | First token | Total | Peak RAM | Completed | Sources | Phone | Shared |
|---|
About the score
Score v1, from 0 to 100: 60% speed (median tokens per second on completed answers of 16 tokens or more, full marks at 20), 25% answers completed and 15% expected sources found. A model that completes fewer than half its answers scores 0. It measures speed and reliability, not whether answers are correct.
The server computes every score from the answers themselves; the app can't send its own. Shared runs carry the phone's model, chipset, cores and CPU features, RAM and app version, never chats, documents, names or location. How the score works.