Skip to content
UXE Award
Featured

vLLM

390 landmarks, one heading.

Visit website

Overall UXE Score

7.87

Assessed by the editors · −0.1 vs season average

vLLM — docs.vllm.ai/en/latest/examples/, captured 28 September 2026 at 1440×900
Recognition
Featured
Recognized
28 Sept 2026
How it got here
Chosen by the editors
Studio
vLLM
Assessed as
AI & Emerging Experience
Platforms
Web

01/Overview

A model-serving engine whose documentation carries 390 landmark elements and exactly one heading, with eight automated violations.

390 landmark elements and one heading. That is the inverse of every other page in this round: the document is divided into hundreds of regions and gives a reader almost no headings to navigate between them. Structure without signposting.

Eight violations: an unnamed ARIA dialog, an unnamed progressbar, one contrast failure, and landmark nesting problems including a duplicated banner. The unnamed progressbar is announced as a progress indicator with no idea what is progressing.

1.35 seconds to largest paint on 1.6 MB over forty requests, with a layout shift of 0.010. A skip link and two reduced-motion rules are present. The performance is good; the document outline is the problem.

Patterns observed

  • No consent dialog on first visit
Client
In-house product
Audience
Teams serving language models at volume
Industry
AI Products
Platforms
Web
Status
Live
Markets
Worldwide
Languages
1 — English

02/Honors

Recognized by the editors.

Recognition

Featured

Meets the UXE standard. Earns a public profile, a badge and a place in the quarterly awards.

Conferred by the editors on 28 Sept 2026 · minimum UXE 7.0

Awards

None. The quarterly awards are decided among projects their makers entered, and nobody entered this one — the editors chose it, assessed it and conferred the recognition above. If vLLM’s team would like it entered and considered for a season, entering is free.

03/Editors’ verdict

In the editors’ words.

“390 landmarks and one heading. Hundreds of regions and nothing to navigate between them.”

The UXE editors

Selected and assessed by the editors

04/Score breakdown

How the editors scored it.

Overall UXE Score−0.1 vs Q3 2026 average
7.87
Strongest criterionPerformance
8.9
Assessed bytheir own scores, signed
The editors
Measured28 Sept 2026
On the live product
UsabilityIAUIInteractionPerformanceA11yLearnability
UXE Fingerprint · scale 5–10 from center
CriterionScorevs avg
  1. Usability

    Can people complete what they came to do — efficiently, confidently and without error?

    Editors’ note390 landmark elements against one heading — structure without signposting.

    8.0−0.2
  2. Information Architecture

    Is content organized, labeled and navigable in the way people actually think?

    7.0−0.9
  3. UI & Consistency

    Is the visual language coherent and purposeful across every screen and state?

    8.2−0.1
  4. Interaction & Feedback

    Does every action respond clearly, promptly and meaningfully?

    7.6−0.2
  5. Performance

    Does it feel fast — perceived and measured — on real devices and networks?

    Editors’ note1.35 s to largest paint on 1.6 MB over forty requests, layout shift of 0.010.

    8.9+1.4
  6. Accessibility

    Can everyone use it, including people relying on assistive technology?

    Editors’ noteAn unnamed ARIA progressbar and an unnamed dialog, plus a duplicated banner landmark. A skip link and two reduced-motion rules are present.

    7.6−0.2
  7. Learnability

    How quickly does a first-time user become a confident one?

    7.8−0.3

Scores are the editors’ own — this product was selected rather than entered, and the verdict is signed. Ticks mark the Q3 2026 average across 161 recognized projects.

05/Screens

2 screens

The product, screen by screen.

vLLM — docs.vllm.ai/, captured 28 September 2026 at 1440×900
01First visit
vLLM — docs.vllm.ai/en/latest/examples/, captured 28 September 2026 at 1440×900
02Second surface

06/Accessibility & performance

Built for everyone, on any device.

Accessibility

Assessed here

Audited by the UXE editors (axe-core 4.10 over the loaded page, plus manual checks), September 2026

7.6

Measured

  • Pinch zoom is not blocked
  • No horizontal scrolling at the assessed viewport
  • 390 landmark regions, and a page language declared as en
  • A skip link was found
  • 2 prefers-reduced-motion rules in the stylesheet

Known gaps

  • 8 automated violations in one pass — 4 serious, 4 moderate
  • ARIA dialog and alertdialog nodes should have an accessible name (1 element, serious)
  • ARIA progressbar nodes must have an accessible name (1 element, serious)
  • Elements must meet minimum color contrast ratio thresholds (1 element, serious)

Performance

3 of 4 within budget

Chrome headless at 1440×900, throttled to Slow 4G (1.6 Mbps down, 150 ms round trip) with 4× CPU slowdown. Three cold loads, each in a fresh browser context with the HTTP cache disabled; the median is reported. 28 September 2026.

8.9

Measured

Performance metrics against budget
MetricMeasuredBudget
Largest Contentful Paint1.35 s (within budget)≤ 2.5 s
First Contentful Paint1.16 s (within budget)≤ 1.8 s
Cumulative Layout Shift0.01 (within budget)≤ 0.10
Page weight1.6 MB (over budget)≤ 1 MB

Within budget Over budget

07/Credits

The people who made it.

Studio

vLLM

The project behind vLLM, an inference and serving engine for language models, developed in public.

Recognized
1
docs.vllm.ai/

Team

Individual credits are published when the team submits them. This profile was compiled by the editors, so it names the company and nobody else.

Client
In-house product
Industry
—

Built with

    →/More from the showcase

    Whole season
    devenv — devenv.sh/blog/, captured 28 September 2026 at 1440×900

    8.87

    UXE

    Official SelectionAssessed Sept 2026
    Starlight — starlight.astro.build/getting-started/, captured 30 September 2026 at 1440×900

    8.83

    UXE

    Official SelectionAssessed Sept 2026
    Josh W. Comeau — www.joshwcomeau.com/blog/, captured 28 September 2026 at 1440×900
    Official SelectionAssessed Sept 2026