---
format: "aidr-story-markdown/v1"
id: "83a753b4160cba93855aa4361255dc1036593f3714dcb4c0d1799bf3d520a424"
canonical_url: "https://aidr.today/83a753b4?lang=en"
title: "vLLM v0.32.0"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-18T06:50:48.000Z"
category: "Open Source"
topics: ["inference","open-source"]
source_urls: []
summary: "vLLM v0.32.0 was released with 814 commits from 363 contributors, including 121 new ones. The release introduces a new POST /v1/systemone endpoint for structured state queries and generic parsers that read checkpoint response templates, eliminating the need for model-specific parsers. It also adds low-precision KV cache support across various hardware platforms, including NVFP4 on SM8x, SM12x, and SM90, FP8 on SM100, and 4-bit ultraquant_4bit on ROCm."
---

# vLLM v0\.32\.0

> [Open the canonical story](<https://aidr.today/83a753b4?lang=en>)

**Published:** 2026-10-18T06:50:48.000Z
**Category:** Open Source
**Topics:** inference, open\-source

## Summary

vLLM v0\.32\.0 was released with 814 commits from 363 contributors, including 121 new ones\. The release introduces a new POST /v1/systemone endpoint for structured state queries and generic parsers that read checkpoint response templates, eliminating the need for model\-specific parsers\. It also adds low\-precision KV cache support across various hardware platforms, including NVFP4 on SM8x, SM12x, and SM90, FP8 on SM100, and 4\-bit ultraquant\_4bit on ROCm\.

## Sources

_No safe source URL was recorded._