Liquid AI d1 3B decides in 8ms on a 3090, crushes OpenAI API (345ms) at Snake
victor_explore · x · 2026-10-09
- A Snake game benchmark compared local vs API model decision latency: Liquid AI's d1 3B running on a single RTX 3090 decided moves in 8ms, making 82 moves per second and eating 86 apples, while OpenAI's API took 345ms round trip per call and only ate 3.
- The reblogger notes most of the gap is network latency, not the model—so it's not a fair capability comparison, though it nicely illustrates the latency edge of small local decision models.
More from Fun
- Feynman was on the original Thinking Machines team; Connection Machine was a proto-GPU — sytelus · 2026-10-09
- Engineers Today Don't Get to Experience: From Pair Programming to Joy — KevinNaughtonJr · 2026-10-09
- mitsuhiko Jokes Python 3.15 Is First Release Where Core Team Suffers 'AI Psychosis' — mitsuhiko · 2026-10-09
- Dev Builds GTA-Style Game With Claude in Hours, Jokingly Recruits a Crowd — Daniel_Farinax · 2026-10-09
- Apple UI screenshot reignites the endless debate: ❎✅ or ✅❎ button order — ssh4net · 2026-10-09
- ARK Analyst's Quantum-Bitcoin Joke: AI Cracks Wallet Math, User Flees Overnight — skorusARK · 2026-10-09