Tencent has released DSpine, a five-layer parallel draft model for speculative decoding with Qwen3.8-27B, licensed under Apache-2.0. Unlike traditional single-token drafters, DSpine generates 16-token blocks in parallel while preserving causal order through predecessor feature passing across layers. Benchmarks in SGLang report 2.7–3.8x end-to-end speedups over standard decoding. The model supports an 8,192-token context and is accompanied by a paper (arXiv:2609.36173) and open-source code on GitHub.