v3.3.0 v3.4.0

Name: inference
Author: xorbitsai

high risk

xorbitsai/inferenceAug 30, 2026 → Sep 11, 202612 days between releases

What's New

Adds Gemma-4, MiniCPM5-2B, and OCI support; requires xoscar 0.10.0 and Python 3.14.

Breaking Changes

Requires xoscar 0.10.0 and Python 3.14 support

New Features

Gemma-4 Transformers backend with batching support
Multi-worker replica scaling
Resumable cache management
OCI URI support
MonkeyOCR integration

Improvements

vLLM native multiprocessing executor routing
Download-only flow for models
Engine registration hooks
TTS streaming playback in UI
Prefill/decode separation

Upgrade Notes

Ensure your environment supports Python 3.14 and xoscar 0.10.0. Review OCI URI configurations if using external model registries. Check multi-worker scaling settings if deploying horizontally.

The Two Releases

v3.3.0

Aug 30, 2026· 2 months ago

This release introduces extensive new features including support for multiple new AI models (Qwen3.8, DeepSeek V4, Ornith-1.5, Kimi-K3, etc.), new audio and image generation models, enhanced router capabilities with token-aware routing and agent-based orchestration, and various UI enhancements.

Full release notes

v3.4.0

Sep 11, 2026· 1 month ago

This release introduces support for new models such as Gemma-4 and MiniCPM5-2B, alongside resumable cache management and multi-worker scaling. It also adds OCR capabilities, OCI URI support, and significant Web UI enhancements including TTS streaming and new language options.

Full release notes