Convai Releases Laya-MLX: Open-Source Decision Engine Achieves 60 Decisions Per Second on Apple Silicon

Convai Innovations has released Laya-MLX, an open-source classification system that performs typed decisions in milliseconds on Apple Silicon hardware, positioning itself as a faster and freely available alternative to proprietary decision engines like Jev. The 421-million-parameter model, built on ModernBERT-large and published under Apache 2.0 licensing, completes single classification tasks in 32.8 milliseconds on standard GPUs and as few as 7–14 milliseconds on M3 Max hardware. A publicly demonstrated Snake game running on an M3 Max achieves approximately 60 decisions per second at full speed, showcasing the system's ability to make rapid, sequential choices without external API calls or cloud infrastructure.