Play the video in the post belowPost on X
- Views
- 24,257
- Likes
- 150
- Reposts
- 23
Read from X on . Counts change daily.
Simplifying AI ran the same Snake game with two models making the same typed decisions, each for a 30-second free run. On the left is a local, open-source 421M-parameter Laya checkpoint. On the right is Jev 1.13.0 called through TypeSafe’s cloud API.
Laya scored 46 with a final length of 52 at 86.5 decisions per second. Jev scored 1 with a length of 7 at 3.2 decisions per second. In a separate test on an M5 Pro, the author measured a median latency of 15.3 ms for Laya running locally against 298.1 ms for Jev over the cloud, “nearly 20x faster”.
The post puts the gap down to the network. Laya runs on the device in about 1 GB of memory, while Jev waits for a cloud round trip at every step.



