vLLM ADOPT

vLLM 0.33 is stable and expands EFA, low-precision paths and diagnostics, but CUDA 13, ROCm and executor changes make it a non-trivial upgrade. Teams needing new models and multi-node serving should adopt it in stages.

Next stepStage the upgrade and regress images, drivers, KV transfer, paused requests and real throughput.

Judgment reviewed 2026-11-03 · Basis: vLLM 0.33 moves stable builds to CUDA 13 and adds EFA serving

RSS

Timeline

2026-11-02Update

vLLM 0.33 moves to CUDA 13 and adds EFA serving and new inference paths

Original story: vLLM 0.33 moves stable builds to CUDA 13 and adds EFA serving

Official · ReferencevLLM 0.33.0 release

2026-10-18Update

vLLM 0.32 tightens security defaults and expands low-precision inference

Original story: vLLM 0.33 moves stable builds to CUDA 13 and adds EFA serving

Official · ReferencevLLM 0.32.0 release

2026-10-05Update

vLLM 0.31 extends fast restart and tightens multimodal request boundaries by default

Original story: vLLM 0.33 moves stable builds to CUDA 13 and adds EFA serving

Official · Release notesvLLM 0.31.0

2026-09-22New

vLLM 0.33 moves stable builds to CUDA 13 and adds EFA serving

Release notesvllm-project/vllm · releases / tag / v0.30.0

Choosing models →Local deployment →