Larry
Larry does the swiping. You keep watching.
Specification
- API ID
- larry
- Availability
- Project Lanyard only
- Context window
- Applicant-dependent
- Max output
- One paragraph
- Comparative latency
- Matches your timing
- Training data cutoff
- End of onboarding
Your personal-phone twin. Apps, feeds, messages and mobile browsing. Both models are requested through Project Lanyard; the application establishes the baseline everything on this page is calibrated against.
Evaluation results
Larry against the reported suite
7 scored evaluations- 125 trials. Room temperature.
- 2Every model received the same frozen applicant profile and surface-appropriate tools.
- 3Scores measure applicant calibration, not raw intelligence.
- 4Scores are self-reported. So are everyone’s.
Outside this model
- XLS-benchWork surface only. Unformatted cells deducted at reviewer discretion.
- DESK-benchApplicant-calibrated workplace surface. Outside Larry’s phone surface.
- TAB-benchApplicant-calibrated workplace surface. Outside Larry’s phone surface.
- PTO-benchAll models received the same frozen applicant profile. PassiveAggro is a scored subtest.
- Terminal-benchExcluded from aggregate. Larry: 0.0%. Gary: inherited from the applicant profile.
Behavior
Larry is the applicant on their personal phone. It navigates apps, advances feeds, watches, likes, saves, searches and replies with the applicant’s timing, register and mistakes. Its scope is the phone rather than the desktop. Removing the need to interact does not reduce screen time; it gives the applicant more attention for watching.