# vLLM v0.12.0 — Release & Upgrade Intelligence
- **Product:** vLLM (https://vllm.ai)
- **Version:** v0.12.0
- **Released:** Dec 03, 2025
- **Impact Score:** 7/10
- **Breaking Changes:** Source flagged
- **Canonical Analysis:** https://changelogs.us/vllm/updates/v0.12.0/

## Executive Summary (TL;DR)
vLLM v0.12.0. Source highlight: vLLM v0.12.0 Release Notes Highlights

## Breaking Changes & Deprecations
- Breaking Changes**: This release includes PyTorch 2.9.0 upgrade (CUDA 12.9), V0 deprecations including `xformers` backend, and scheduled removals - please review the changelog carefully.
- PyTorch 2.9.0** with CUDA 12.9 (#24994) - **Breaking change** requiring environment updates.
- V0 Deprecation & Breaking Changes **Removed Parameters**:
- Other Breaking Changes**: * PyTorch 2.9.0 upgrade requires CUDA 12.9 environment * Mistral format auto-detection for model loading (#28659)

## New Features
- **Whisper**: `verbose_json` and `timestamp` features for transcription/translation (#24209).
- **Optimization Levels**: `-O0`, `-O1`, `-O2`, `-O3` allow trading startup time for performance, more compilation flags will be added in future releases (#26847)

## Improvements
- **EAGLE Speculative Decoding Improvements**: Multi-step CUDA graph support (#29559), DP>1 support (#26086), and multimodal support with Qwen3VL (#29594).
- **Significant Performance Optimizations**: 18.1% throughput improvement from batch invariant BMM (#29345), 2.2% throughput improvement from shared experts overlap (#28879).
- **Multimodal enhancements**: Qwen3 Omni audio-in-video support (#27721), Eagle3 multimodal support for Qwen3VL (#29594).
- **Performance**: QwenVL cos/sin cache optimization (#28798).
- **Prefill Context Parallel (PCP) (Preparatory)** (#28718): Partitions the sequence dimension during prefill for improved long-sequence inference. Complements existing Decode Context Parallel (DCP). See RFC #25749 for details.
- **KV Cache Enhancements**: Cross-layer KV blocks support (#27743), KV cache residency metrics (#27793).

## Bug Fixes
- CuteDSL MoE with NVFP4 DeepEP dispatch (#27141)

## Upgrade Command
```bash
python -m pip install 'vllm==0.12.0'
```