Real-time Doom demo powered by Jev
A game demo shows real-time decisions. Diogo says it makes about ten Jev calls per second; the post includes a 2:43 video.
Jev-Omni locally runs Chrome Dino on M4 Max
Erik Kokalj demonstrates Jev-Omni, a 12B decision model (Gemma 4 finetune) converted to MLX 8-bit, running fully local on M4 Max at ~100 ms. It outputs P(jump/duck/nothing) for a road slice to play Chrome Dino, no text generation.
Jev-Omni playing Chrome Dino, fully local on an M4 Max 🦖 12B decision model (Gemma 4 finetune) converted to MLX 8-bit, running on M4 Max at ~100 ms. No text generation, it just returns P(jump/duck/nothing) for a slice of road cropped ahead of the dino to cover latency scored
· 0 likesOpen on X
More like this
Jev Plays Doom in Real Time Like an Actual Player
Jev plays Doom in real time and looks like an actual player. It is 20 to 200x faster, 40 to 400x cheaper, with free output tokens. This suggests game opponents could be real-time AI, or a replacement for human assistants.
TypeSafe's new model Jev plays Super Mario Bros.
A user shares TypeSafe's new model Jev playing Super Mario Bros., praising its fast inference and structured outputs for real-time use cases.
