vLLM vs Xinference
Short answer
vLLM: High-throughput LLM inference server for GPUs with an OpenAI-compatible API. Xinference: Model-serving platform for LLM, embedding, rerank, speech and image models behind one OpenAI-compatible API. Both are in our Local model runtimes category.
Choose vLLM if you want: larger community (95.5k vs 9.6k GitHub stars).
Choose Xinference if you want: no GPU server needed.
Xinference publishes no official RAM figure, so we cannot compare hosting cost directly.
Last updated · computed from each project's official docs and live GitHub data · sources
vLLM vs Xinference side by side
| vLLM | Xinference | |
|---|---|---|
| Category | Local model runtimes | Local model runtimes |
| What it is | High-throughput LLM inference server for GPUs with an OpenAI-compatible API. | Model-serving platform for LLM, embedding, rerank, speech and image models behind one OpenAI-compatible API. |
| License | Apache-2.0 | Apache-2.0 |
| License type | Permissive | Permissive |
| License conditions | Permissive. Internal and commercial use allowed. Models keep their own licenses. | Permissive. Internal and commercial use allowed; keep the Apache-2.0 license and notices. Xinference Enterprise is a separate commercial edition. Models keep their own licenses. |
| Status | Active | Active |
| GitHub stars | 95.5k | 9.6k |
| Last commit | Oct 26, 2026 | Oct 26, 2026 |
| Latest release | v0.32.0 | v4.1.0 |
| Minimum hardware | Not published | Not published |
| Recommended hardware | Not published | Not published |
| RAM measured by us | Not tested yet | Not tested yet |
| GPU | NVIDIA: compute capability 7.5 or higher (e.g. T4, RTX 20xx, A100, L4, H100, B200). | The default Docker image requires an NVIDIA GPU (host CUDA 13.0 or newer, driver 580 or newer, NVIDIA Container Toolkit). For CPU, use the image tag with a -cpu suffix; pip installs on Apple Silicon can use the MLX engine. |
| Est. server cost | GPU server (not estimated) | Not estimated |
| Deploy method | pip / uv (or Docker) | Docker |
| Known gotchas | 1 listed | 4 listed |
| Telemetry | On by default | Not documented |
| Calls home by default | Not documented | Yes, by default |
| SSO (OIDC / SAML / LDAP) | Not documented | Free (OIDC free) |
| Audit log | Not documented | Yes, free |
| RBAC | Not documented | Yes, free |
| Details | Full vLLM guide → | Full Xinference guide → |
Key differences
- GPU: vLLM: needed for practical use. Xinference: optional / see notes.
- Est. server cost: vLLM: GPU server (not estimated). Xinference: not estimated (no official RAM figure).
- Deploy: vLLM: pip / uv (or Docker) · pip / uv (or Docker). Xinference: Docker · single container.
- GitHub stars: vLLM: 95.5k. Xinference: 9.6k.
- Latest release: vLLM: v0.32.0 (Oct 18, 2026). Xinference: v4.1.0 (Oct 25, 2026).
Same for both: Category: Local model runtimes; License: Apache-2.0 (Permissive); Official hardware: not published; Status: Active; Last commit: Oct 26, 2026.
vLLM vs Xinference: FAQ
vLLM or Xinference: which should a small business choose?
Choose vLLM if you want: larger community (95.5k vs 9.6k GitHub stars). Choose Xinference if you want: no GPU server needed. Xinference publishes no official RAM figure, so we cannot compare hosting cost directly.
Which is cheaper to self-host, vLLM or Xinference?
vLLM: GPU server (not estimated). Xinference: not estimated (no official RAM figure). Our estimate is the cheapest Hetzner Cloud plan that meets each project's official RAM and vCPU figure; it covers the server only.
How much hardware do vLLM and Xinference need?
vLLM: not published. Xinference: not published. vLLM needs a GPU for practical use.
What is the license difference between vLLM and Xinference?
vLLM: Apache-2.0 (Permissive). Permissive. Internal and commercial use allowed. Models keep their own licenses. Xinference: Apache-2.0 (Permissive). Permissive. Internal and commercial use allowed; keep the Apache-2.0 license and notices. Xinference Enterprise is a separate commercial edition. Models keep their own licenses. This is a summary, not legal advice.
Is vLLM or Xinference more actively maintained?
vLLM: Active, last commit Oct 26, 2026. Xinference: Active, last commit Oct 26, 2026. GitHub data refreshed Oct 26, 2026.
Which is easier to deploy, vLLM or Xinference?
vLLM installs via pip / uv (or Docker) (pip / uv (or Docker)); Xinference via Docker (single container). vLLM lists 1 known gotcha and Xinference lists 4; see each app page.
Related
- vLLM self-hosting guide · Xinference self-hosting guide
- All vLLM alternatives, ranked
- All Xinference alternatives, ranked
- GPUStack vs vLLM
- GPUStack vs Xinference
- llama.cpp vs vLLM
- llama.cpp vs Xinference
- LocalAI vs vLLM
- LocalAI vs Xinference
- Ollama vs vLLM
- Ollama vs Xinference
- Add a third app in the interactive comparison
Sources
- vLLM GitHub repository: github.com/vllm-project/vllm
- vLLM documentation, vLLM hardware requirements: docs.vllm.ai/en/latest/getting_started/installation/gpu/
- vLLM license: github.com/vllm-project/vllm/blob/main/LICENSE
- vLLM deploy instructions: github.com/vllm-project/vllm#readme
- Xinference GitHub repository: github.com/xorbitsai/inference
- Xinference documentation, Xinference hardware requirements, Xinference deploy instructions: inference.readthedocs.io/en/latest/getting_started/using_docker_image.html
- Xinference license: github.com/xorbitsai/inference/blob/main/LICENSE
- Hetzner Cloud pricing used for cost estimates (read 2026-10-26): www.hetzner.com/cloud/regular-performance
- GitHub REST API (stars, commits, releases): docs.github.com/en/rest
Every value is generated from the same verified data as the app pages. Spotted something wrong? Tell us.