Slotstream Brings a 105 GB Qwen Model to Macs with 16 GB of Memory

Slotstream, a local inference engine for Apple Silicon, can run Qwen3.8-Flash-Next on Macs with 16 to 64 GB of memory by streaming model data from an SSD. The project says inference runs on the Mac and works offline after the model has been downloaded. Its project documentation describes a Swift engine built on Apple’s MLX and Metal that keeps frequently needed model experts in memory while reading others from storage.