Release Notes · July 2026

Xinference 3.0

The most significant upgrade since launch. A brand-new frontend, enterprise authentication, modern deployment — rebuilt from the ground up.

🎨 Next.js UI 🔐 Auth System 🐳 Docker Compose 🧩 vLLM BCE
⚠️

Breaking Changes — Migration Guide Required

Version 3.0 introduces several breaking changes to default behaviors. If upgrading from 2.x, please read the migration guide carefully before proceeding.

Read the Migration Guide →

What's New

Highlights

🎨

Next.js Web UI

The new Next.js frontend becomes the default, delivering a completely redesigned interface and interaction experience. React and Gradio UIs are officially retired.

  • • API Key management dashboard
  • • User management & security settings
  • • Audit center with full traceability
🔐

Enterprise Auth

Database-backed authentication is now enabled by default, bringing production-grade security to every deployment.

  • • API Key lifecycle management
  • • Password change & first-login enforcement
  • • Refresh Token & offline password reset
🐳

Docker Compose

A production-ready Docker Compose deployment stack is now available, supporting fully offline environments with a built-in PyPI mirror.

  • • GPU slim images (no pre-installed engines)
  • • Offline deployment support
  • • Built-in PyPI mirror proxy
🧩

Embedding & llama.cpp

Expanded inference capabilities with new embedding support and smarter llama.cpp integration.

  • • vLLM BCE Embedding support
  • • Auto-install CUDA-matched xllamacpp
  • • GPU memory display for running models

Heads Up

Breaking Changes

🎨

Default Frontend → Next.js

React & Gradio UIs are no longer the default; Next.js is now the primary web interface.

🔐

Database Auth Enabled by Default

Fresh installs come with the new authentication system active out of the box.

🐳

Docker Deployment Upgraded

New Docker Compose workflow replaces previous single-container approach.

📦

GPU Slim Images

GPU Docker images no longer bundle inference engines; install on demand for smaller footprints.

Upgrading from 2.x? Configuration migration is required.

Migration Guide →

Open Source

Community Edition

Installation

pip

pip install 'xinference==3.0.0'

Docker

docker pull xprobe/xinference:latest

or use the new Docker Compose deployment

✨ New Features

  • ▸ Next.js as default Web UI
  • ▸ API Key management page
  • ▸ User management page
  • ▸ Security Settings page
  • ▸ Audit Center page
  • ▸ Database auth enabled by default
  • ▸ Offline password reset
  • ▸ Offline Docker Compose deployment
  • ▸ GPU Docker Slim Image support
  • ▸ Running Models page shows GPU memory
  • ▸ BCE Embedding support (vLLM)
  • ▸ Auto-install CUDA-matched xllamacpp for llama.cpp

🛠 Enhancements

  • ▸ Updated model configs (Embedding / Audio / Rerank / LLM)
  • ▸ Updated CogVideoX, HunyuanVideo model configs
  • ▸ GPU CI & Metal CI continuous improvements

🐞 Bug Fixes

  • ▸ FlashInfer Blackwell support
  • ▸ DeepSeek-V4 Tool Call
  • ▸ Fun-ASR dependency issues
  • ▸ Multi-worker shard launch
  • ▸ Llama.cpp multimodal mmproj path
  • ▸ OAuth2 Refresh Token revocation
  • ▸ PyTorch Rerank GPU memory release
  • ▸ Virtualenv multi-engine matching
  • ▸ DeepSeek OCR torchvision import
  • ▸ Offline Package build dependencies

📚 Documentation Updates

Enterprise

Enterprise Edition

🚀

Built on 3.0 Core

Fully upgraded on the Xinference 3.0 open-source foundation with enterprise-grade authentication, a redesigned admin console, and enhanced deployment capabilities.

🌏

Vietnamese Language

New Vietnamese language support added to meet the needs of more international enterprise deployments.

🛡

Stability & Security

Extensive stability, security, and enterprise capability improvements across the platform.