vLLM 0.29 makes Model Runner V2 the default
vLLM 0.29 made Model Runner V2 the default for most models, added Hy4, Qwen 3.8 Flash Next, Granite SWA and Kimi K3 variants, and expanded CUDA-graph, KV-cache, multimodal, LoRA and distributed-inference paths. Signed wheels and container images cover CUDA, ROCm, CPU and XPU.
Why it matters: A default execution-engine change can alter memory, latency, compatibility and failure behaviour even when application code and model IDs stay fixed, while the broader model matrix expands deployment options.