Maya-XS runs DeepSeek‑V4.1‑Flash entirely in a 128 GB PC memory, keeping all 384 experts resident on the GPU. The model uses tensor‑parallelism across multiple GPUs and delivers much faster drafting on a single GPU.