A text-in, text-out model published by sakamakismile under an open Apache 2.0 license, making weights freely available for inspection and deployment. It belongs to the Qwen3.8 family and uses NVFP4 quantization with multi-token prediction, which typically trades some precision for faster inference. Concrete capability details beyond its architecture lineage are limited.