Games with Laya

The port announcement: Laya on MLX, under 1 GB RAM, choosing the snake’s next move 60 times a second on a local M3 Max.

Atomic.chat ran Laya on a 16 GB MacBook Air against Jev over the API on a Grok-built Tetris. Local won on speed: 11x faster decisions.

Viraj Bhartiya wired both models into the Chrome T-Rex game: Laya local on MLX, Jev over the API, same track and physics, side-by-side latency. Code in virajbhartiya/laya-vs-jev.

Tony Dinh ran laya-mlx on a MacBook against Jev in real-time Tetris. Laya decided far faster but stacked itself to the ceiling; Jev won every round.

Simplifying AI ran the identical Snake and typed decisions for 30 s on each model. Laya (421M, local) scored 46 at 86.5 decisions/s; Jev 1.13.0 over the API scored 1 at 3.2/s.

A maze demo on a base M5 Mac where local Laya answered in about 19 ms median against roughly 300 ms for the hosted Jev API. Shows the latency gap that on-device decision models open up.

Tom Siwik’s point: the model matters less than how you define the decision space and batch. Jev wins networked; Laya-mlx and his own JAX build tie locally.

Tobias built his own Snake benchmark and ran the Jev API against Laya-MLX and Kev-4B on a low-RAM MacBook. Honest take: Jev wins on quality, locals win on speed.

Shantanu Goel trained Laya for ~24 hours to play FreeDoom E1M1 from text-only perception. Jev played well zero-shot; the Laya base model did not. Honest negative result.

AmirTaha wrote a Snake game that feeds fruit, head and border positions to a local Laya model and takes the next direction from it. Persian post, code in AmirTahaMim/LayaSnakeGame.

Hope put both models in a 1v1 Doom deathmatch, first to five kills. Jev's calls were sharper and needed a third of the corrections; Laya decided twice as often.

An experimental mod that lets the open-weight Laya model manage a RimWorld colony autonomously on the player's own machine.

One typed question per frame; the other paddle is three lines of arithmetic as a control. Physics in WebAssembly, decisions in native Laya over POST /decide.

Ports the laya-mlx Snake demo to the DGX Spark GB10, showing the open Laya decision model working on Nvidia hardware.

Mgear runs the receptron/laya-onnx weights in Unity through onnxruntime-unity and plays tic-tac-toe against the model.

A Snake game whose moves all come from a fine-tuned copy of Laya, showing a decision model acting in a real-time control loop.

A modern Sudoku game where the Laya decision model picks moves with about 35 ms of GPU latency.

Two decision models fight a real Doom deathmatch with the same state, questions and shield: open local Laya against hosted Jev.

Ahmed made the 322M multilingual checkpoint drive every paddle move in Breakout, locally on a Windows laptop with zero cloud. Code in Eng-Ahmd/laya-breakout.

Daniel F built a fighting game with 21 world-leader-inspired fighters, signature moves, combos and replays. The opponent can be driven by Laya in the browser.

A duel bot for Ultima Online that lets local Laya and cloud Jev fight, with a live monitor comparing their decisions.

A Rubik's-cube agent that combines 3D state, temporal memory and calibrated decision probabilities from Laya and Jev to choose each turn.

Runs the open 421M Laya model as the controller in a real-time Doom loop, a response to TypeSafe's Jev Doom demo.

A Rich-based terminal Tetris built so a local Laya MLX model can play it on Apple silicon.