Open-weights Laya plays Tetris locally, deciding 11x faster than cloud-based Jev on a MacBook Air
alexcovo_eth · x · 2026-09-22
Atomic Chat demoed Laya, an open-weights System One-style classifier model (similar to Jev), playing Tetris locally on a 16GB MacBook Air and making decisions 11x faster than cloud-based Jev. The model was ported to MLX with optimizations, reportedly 50x faster with under 1GB of memory — hitting 60 decisions/sec playing Snake on an M3 Max. The post also promotes Atomic Chat, an open-source local AI chat client across Mac, Windows, and mobile.
Related event: Open-Source Laya Ported to Apple MLX for 7ms On-Device Decisions(3 posts)→
More from Infra
- HBM4 yield issues at SK Hynix and Micron delay NVIDIA Rubin ramp, KYEC cuts Q4 guidance — zephyr_z9 · 2026-11-06
- AI's Next Architecture Question: Where Inference Should Actually Run — DavidLinthicum · 2026-11-06
- Alphabet's 13F reveals ~3M Qualcomm shares worth $548M, sparking chip partnership speculation — zephyr_z9 · 2026-11-06
- Floci: open-source local AWS emulator runs 120+ services in 24ms, free LocalStack alternative — thisguyknowsai · 2026-11-06
- NVIDIA puts AI servers in telecom sites: ~20 operators join AI-RAN push — shashib · 2026-11-06
- Sharp Tech Goes Video: Data Center Votes, Big Tech Implications, Google's AI Response — Stratechery · 2026-11-06