vLLM
docs.vllm.ai · Inference & Hosting
High throughput LLM serving engine built around PagedAttention, with an OpenAI compatible server.
vLLM is an open source (Apache 2.0) engine for serving large language models on your own GPUs, started at UC Berkeley and known for PagedAttention, which manages the attention key value cache in pages to fit more concurrent requests. Its vllm serve command exposes an OpenAI compatible HTTP server, so agent frameworks, LiteLLM and the OpenAI SDKs can call a self hosted model by changing the base URL.
The latest release on GitHub was v0.32.0 (October 18, 2026). vLLM itself is free; you pay for the hardware it runs on. Use it when you need throughput for many users on your own infrastructure; for a single developer machine, Ollama or LM Studio are simpler, and SGLang is the closest open source alternative for production serving.
vLLM is one of 10 inference and hosting tools on Drydock. See all inference and hosting tools or how we check facts.
