

vLLM v0.34.0 landed on November 16, 2026, continuing the v0.33+ serving architecture that standardized on Model Runner V2 and CUDA 13.0 builds (requiring R580 or newer NVIDIA drivers).
Upgrade considerations
Clusters still pinned to CUDA 12.9 remain on the v0.32.0 branch. On supported Blackwell and AMD MI355X deployments, the v0.33/v0.34 runner also enables extensible KV-cache sizing post-warmup by default.
EXPLORE THE SOURCE
Projects in this story
GitHub repositories
- vllm-project/vllm ↗View project details
Original reporting; primary project releases reviewed November 16, 2026. No independent hands-on benchmark claimed.



