# Bootstrap 6: 15 years of history, Bootstrap 5 today and the big rebuild in the alpha

> Bootstrap from Twitter Blueprint to 6.0.0-alpha.1: the history, where 5.3.9 stands, what's new in v6 and how much the team works with AI.

Source: https://www.jpkc.com/db/en/blog/bootstrap-6/

Bootstrap has been with me since the very first version. I've used it in countless projects, my own as well as projects for others, from the responsive add-on stylesheet through mobile first and Flexbox to the farewell to jQuery (a JavaScript library Bootstrap relied on up to version 4). Traces of it are still in my work today; the main site [jpkc.com](https://www.jpkc.com/en/), for example, still has some Bootstrap 4 heritage in its reset. In the [article on cascade layers](https://www.jpkc.com/db/en/blog/css-cascade-layers/) I counted through Bootstrap 5.3.8 and showed how you can put the framework into a layer of your own (a named priority level in CSS) after the fact, because it doesn't ship a single one itself.

That's exactly what Bootstrap 6 changes, and not only that. The first alpha rebuilds the framework more fundamentally than any version since Bootstrap 4, and I consider the new features very innovative. At the same time it's an alpha, the browser floor is high, and the switch shifts values without you noticing at first glance. So this article gives you everything: the history since 2010, the state of Bootstrap 5, what's new in v6 along with its limits, and a look at how heavily the Bootstrap team now works with AI.

## The short version

| Question | Answer |
| --- | --- |
| **What is Bootstrap?** | A free CSS framework with a grid, components, utility classes and JavaScript plugins, released at Twitter in 2011, today under the MIT license |
| **How widespread is it?** | According to W3Techs on 12.9% of all websites surveyed, and clearly ahead with a 74.8% share among CSS frameworks. On npm around 7.5 million downloads per week |
| **What's stable?** | Bootstrap 5.3.9 from 11 November 2026, on npm under `latest`. There's no end-of-life date for v5 yet |
| **What is Bootstrap 6?** | Version 6.0.0-alpha.1 from 8 October 2026, on npm under `next`. An alpha, explicitly still subject to change |
| **What's new?** | Cascade layers, colors in oklch, dark mode following the system setting, Sass modules, prefix classes like `md:d-none`, components built on `<dialog>` and `<details>`, TypeScript, ES modules only, 20 instead of 12 plugins |
| **What does switching cost?** | Chrome and Edge 130, Firefox 132 and Safari 18 as the minimum, new values for spacing and breakpoints, renamed components, considerably larger files |
| **What role does AI play?** | `llms.txt`, 11 agent skills, a "For Agents" section on the home page, triage with Claude Code agents and many commits with Claude as co-author. There's no MCP server |

All of this refers to Bootstrap 5.3.9 and 6.0.0-alpha.1, as of 12 November 2026. "According to the docs" means read in the official documentation. "Counted" means counted in the npm packages 5.3.8, 5.3.9 and 6.0.0-alpha.1, without running their scripts. The code examples come from the docs, and a few are put together from the docs and the package. They haven't been tested in a project of my own. Third-party opinions and measurements appear with their source, and my own assessments are marked as such.

## What Bootstrap is

A CSS framework is a ready-made collection of stylesheets, often with a bit of JavaScript, that takes the groundwork of a website off your hands. Bootstrap provides four building blocks: a grid (a layout of rows and columns that adapts to the screen width), components such as buttons, navigation bars, forms and dialogs, utility classes (small classes with exactly one job, such as `.mt-0` for "no top margin") and JavaScript plugins for everything that moves or expands. You write HTML with the matching classes and get a consistent, responsive interface, meaning one that works on a smartphone just as well as on a large monitor.

For professionals, Bootstrap is more than a construction kit. Through Sass (a CSS preprocessor with variables, functions and modules) and CSS custom properties (variables right in the browser) you adjust colors, spacing and components, and with the utility API you generate your own utility classes from a configuration. The [official home page](https://getbootstrap.com/) now calls Bootstrap "the open source design system for everyone", in other words a design system (a defined set of design rules, tokens, i.e. named base values such as colors and spacing, and components) rather than just a framework. What makes up a design system as a file is something I described in the [article on DESIGN.md](https://www.jpkc.com/db/en/blog/design-md/).

After 15 years, its reach is unbroken. [W3Techs](https://w3techs.com/technologies/overview/css_framework), with data from 12 November 2026, sees Bootstrap on 12.9% of all websites surveyed and with a 74.8% share among websites with a detectable CSS framework. Second place goes to Animate.css, with 7.2% of all websites and a 41.7% share among frameworks (the shares overlap because many websites include several frameworks). Foundation reaches 0.4% of all websites and a 2.3% share, Tailwind CSS (a framework built almost entirely on utility classes) 0.3% and 1.8%. On npm, the central package registry for JavaScript, the `bootstrap` package had 7,511,430 downloads in the week from 5 to 11 November 2026 according to [npm statistics](https://api.npmjs.org/downloads/point/2026-11-05:2026-11-11/bootstrap), about 51% more than in the same period of 2025. The [twbs/bootstrap](https://github.com/twbs/bootstrap) repository counts 175,049 stars and 79,124 forks.

At first glance these numbers contradict each other. The `tailwindcss` package had over 176 million downloads in the same week, 23 times as many, while W3Techs sees Tailwind on only 0.3% of all websites. One explanation that I find plausible but can't prove: npm counts every installation in a build pipeline, and W3Techs detects compiled utility CSS less reliably than the typical Bootstrap classes. So the two numbers measure different things, and neither on its own works as proof of "the most used framework".

## The history: from Twitter Blueprint to Bootstrap 5

If you want to understand Bootstrap 6, it helps to know where it comes from. Every major version responded to what the web demanded at the time, and that's exactly how the alpha reads too.

### 2010 and 2011: an internal tool at Twitter

According to the [official project overview](https://getbootstrap.com/docs/5.3/about/overview/), Bootstrap was created at Twitter in mid-2010, built by Mark Otto (@mdo) and Jacob Thornton (@fat). Before its release it was called "Twitter Blueprint". Mark Otto described the beginnings in January 2012 at [A List Apart](https://alistapart.com/article/building-twitter-bootstrap/): a small group wanted to unify the internal analytics and admin tools, and the goal then grew into a toolkit "for anyone to use within Twitter, and beyond". Its first real test was Twitter's first Hack Week. The [English Wikipedia](https://en.wikipedia.org/wiki/Bootstrap_(front-end_framework)) names as the starting point the various libraries used for interface development at Twitter, which led to inconsistencies and a high maintenance burden.

Bootstrap was released on Friday, 19 August 2011. The first commits in the GitHub repository, back then still under `twitter/bootstrap`, date from the last week of April 2011 according to the [contributors statistics](https://github.com/twbs/bootstrap/graphs/contributors). A small curiosity on the side: in 2011 the package name `bootstrap` on npm initially belonged to an unrelated Node project, and according to the registry the first official Bootstrap version on npm is only 3.1.1 from 9 June 2014.

### Bootstrap 2: responsive as an add-on

Bootstrap 2 came out on 31 January 2012. The [project overview](https://getbootstrap.com/docs/5.3/about/overview/) describes the most important step like this: "With Bootstrap 2, we added responsive functionality to the entire framework as an optional stylesheet." So back then responsive web design was an add-on you had to include deliberately. The original blog post on Bootstrap 2 could no longer be retrieved, so I'm leaving out details beyond this core statement.

### Leaving Twitter and the move to twbs

In September 2012 both founders left Twitter. In the blog post ["Onward"](https://blog.getbootstrap.com/2012/09/29/onward/) from 29 September 2012, Mark Otto wrote: "Despite us leaving Twitter to go our separate ways, we'll both be continuing our work on the project." According to [TechCrunch](https://techcrunch.com/2012/09/29/bootstraps-maintainers-leave-twitter-for-github-and-obvious-corp-will-move-it-into-its-own-organization/), Otto went to GitHub and Thornton to Obvious Corp., and the step had "nothing to do with a disagreement about Bootstrap". The move into an organization of its own was completed with the [release candidate of Bootstrap 3](https://blog.getbootstrap.com/2013/07/27/bootstrap-3-rc1/) in July 2013: "Bootstrap has moved to @twbs/bootstrap." The stars and the entire Git history moved along, "New digs, same code."

In my view, that's one of the reasons Bootstrap still holds up today: the project never depended on a corporation's strategy, but on its maintainers.

### Bootstrap 3: mobile first and flat

Bootstrap 3 came out on 19 August 2013, two years to the day after the first release. The [release post](https://blog.getbootstrap.com/2013/08/19/bootstrap-3-released/) sums up the direction in two sentences: "With v3, we've gone flat." and "Mobile first and always responsive!" Mobile first means the base layout is designed for small screens, and larger screens get extensions. On top of that came Glyphicons, now as an icon font instead of images, a reworked Customizer that now compiled right in the browser, `box-sizing: border-box` as the default and a grid with four tiers. Internet Explorer 7 and Firefox 3.6 were dropped.

After that, 19 August almost became a ritual: the first alpha of Bootstrap 4 also came out on that date, two years later.

### The switch to the MIT license

Less visible, but decisive for many projects, was the change of license. Bootstrap was originally under the Apache License 2.0, which isn't compatible with GPLv2. That was a problem for Drupal, among others. [Issue #2054](https://github.com/twbs/bootstrap/issues/2054), asking for MIT, was opened in February 2012 and closed the next day with "Changing the license of Bootstrap isn't practical". In October 2012 the issue was reopened, and in 2013 the team collected the contributors' consent. Contributions from 3.0.1 onwards were dual-licensed, and on 19 December 2013 Otto reported "the MIT relicensing has been completed". Commits whose authors hadn't agreed were reverted and rewritten.

Since [Bootstrap 3.1.0](https://blog.getbootstrap.com/2014/01/30/bootstrap-3-1-0-released/) from 30 January 2014, this applies: "As of v3.1, Bootstrap ships under the MIT license to allow as many people to utilize Bootstrap as possible." For you that still means today: you may use Bootstrap with practically no strings attached, including in client projects and commercial software, as long as the license notice is kept.

### Bootstrap 4: the long haul

Bootstrap 4 is the version with the longest run-up. The post on [3.3.0](https://blog.getbootstrap.com/2014/10/29/bootstrap-3-3-0-released/) from October 2014 said: "While the first alpha is a couple weeks off …". In fact the [alpha](https://blog.getbootstrap.com/2015/08/19/bootstrap-4-alpha/) only arrived on 19 August 2015, and the final version on 18 January 2018. The [release post](https://blog.getbootstrap.com/2018/01/18/bootstrap-4/) mentions "roughly 6,000 commits to v4" and openly admits: "It's literally taken us years to do it".

In terms of content, v4 was a rebuild on a broad front: Sass instead of Less as the preprocessor, Flexbox (a CSS layout model for flexible rows and columns), cards replacing wells, thumbnails and panels, and Reboot as the new base stylesheet built on Normalize.css. The units switched from pixels to rem and em. The last v4 is 4.6.2 from 19 July 2022, and since 1 January 2023 Bootstrap 4 has been at the end of its life (EOL for short) according to the [official support table](https://github.com/twbs/release). The open source project no longer ships updates for v3 and v4. For security updates, the Bootstrap home page points to the commercial [long-term support from HeroDevs](https://www.herodevs.com/support/nes-bootstrap).

Bootstrap 3, by the way, got one late straggler: [version 3.4.0](https://blog.getbootstrap.com/2018/12/13/bootstrap-3-4-0/) from December 2018 brought, among other things, XSS fixes (protection against injected script) for several plugins, and the project's Open Collective page launched with that release. The last v3 is 3.4.1 from February 2019, EOL since 24 July 2019. That EOL versions no longer get security updates isn't theoretical, either: for 3.x, the GitHub Advisory Database lists, for example, [CVE-2024-6485](https://github.com/advisories/GHSA-vxmc-5x29-h64v), an XSS vulnerability in the button plugin, with no patched version. If you still run v3 or v4, that belongs in your planning.

### Bootstrap 5: no jQuery, no Internet Explorer

The [first alpha of Bootstrap 5](https://blog.getbootstrap.com/2020/06/16/bootstrap-5-alpha/) came out in June 2020, the [final version](https://blog.getbootstrap.com/2021/05/05/bootstrap-5/) on 5 May 2021. Under JavaScript, the release post says simply "No more jQuery!", and according to the alpha post Internet Explorer was dropped as well. New were CSS custom properties in some components (across the board only from 5.2), the utilities API, an `xxl` breakpoint (another screen width at which the layout switches), offcanvas (a panel that slides in from the side), a new accordion (collapsible sections), floating labels and experimental support for right-to-left languages (RTL). For that, classes switched to logical names: `.ml-3` (margin-left) became `.ms-3` (margin-start), because "left" means "right" in Arabic or Hebrew. The [project overview](https://getbootstrap.com/docs/5.3/about/overview/) describes it in deliberately modest terms: "Bootstrap 5, focuses on improving v4's codebase with as few major breaking changes as possible."

### Every version mirrors its web

| Version | Final | Defining change |
| --- | --- | --- |
| 1 | 19 Aug 2011 | A unified toolkit from Twitter |
| 2 | 31 Jan 2012 | Responsive as an optional stylesheet |
| 3 | 19 Aug 2013 | Mobile first, flat design, its own `twbs` organization |
| 3.1 | 30 Jan 2014 | MIT instead of Apache 2.0 |
| 4 | 18 Jan 2018 | Sass, Flexbox, cards, Reboot |
| 5 | 5 May 2021 | No jQuery and IE, custom properties, utilities API, RTL |
| 5.3 | 30 May 2023 | Color modes and dark mode |
| 6 | Alpha since 8 Oct 2026 | Cascade layers, modern CSS, native HTML, ESM, AI resources |

When I look at this list, I see a project that rarely ran out in front, but reliably rebuilt itself once a technique had reached the mainstream. Bootstrap 6 breaks with that pattern to some extent, and that's the core of what I find so exciting about the alpha.

## Who is behind Bootstrap

For one of the most widely used frontend projects in the world, the team is astonishingly small. The [team page](https://getbootstrap.com/docs/5.3/about/team/) lists twelve names, led by the two founders, and describes the project as "maintained by the founding team and a small group of invaluable core contributors". According to the [GitHub contributors statistics](https://github.com/twbs/bootstrap/graphs/contributors), Mark Otto (mdo), with over 15,000 commits, is still by far the main author, followed by cvrebert with just over 3,500. The most recent release notes mainly feature Mark Otto and Julien Déramond.

The budget is even smaller. On [Open Collective](https://opencollective.com/bootstrap), as of the reference date, there's a balance of US$54,645.73 out of total net income of US$63,580.31 (after fees; gross US$72,877.73), and the estimated annual budget is only US$5,226.70. The biggest contributor is GitHub Sponsors with US$34,119.05. So little money flows through Open Collective into working time; a large part sits in the account. That's only one channel, though: some people support the maintainers directly through their GitHub profiles, and the docs pages carry Carbon Ads advertising, whose income doesn't show up there. Mark Otto himself writes in the alpha post that Bootstrap had been quiet for a long time while he focused on Pierre (according to the post mainly on work on diffs, trees and code storage). My conclusion from this: to this day Bootstrap lives mainly from the commitment of a few people, and in my view that also explains why the team embraces AI agents so openly. More on that later.

On security, the project has visibly stepped up in recent years. Mark Otto and Julien Déramond took part in the [GitHub Secure Open Source Fund](https://blog.getbootstrap.com/2025/08/10/github-secure-open-source-fund/) in 2025. Among the outcomes were an SBOM analysis (a bill of materials listing all dependencies), private vulnerability reporting, an incident response plan (a playbook for emergencies) and GitHub Actions pinned to fixed commit hashes. With [PR #41973](https://github.com/twbs/bootstrap/pull/41973), a pull request, i.e. a proposed change to the repository, the team removed the link to a Chinese translation of the docs at the end of 2025, because the domain was "previously involved in the `polyfill.io` supply chain attack".

## Bootstrap 5 today: stable, maintained, with legacy baggage

If you start a project with Bootstrap today and simply type `npm install bootstrap`, you get Bootstrap 5. That's the stable line, and it's still being maintained. At the same time it shows its age in places you notice in everyday work.

### 5.2 and 5.3: the quiet big step

Two minor versions changed Bootstrap 5 more than their numbers suggest. [Bootstrap 5.2.0](https://blog.getbootstrap.com/2022/07/19/bootstrap-5-2-0/) from July 2022 brought CSS variables to almost all components: "Nearly all our components now have CSS variables for real real-time customization, easier theming". With that you can, for example, recolor a single button by overriding a variable on the element, without recompiling Sass. A responsive offcanvas and the `_maps.scss` file came along too.

[Bootstrap 5.3.0](https://blog.getbootstrap.com/2023/05/30/bootstrap-5-3-0/) from May 2023 introduced color modes, meaning switchable color schemes, starting with dark mode. According to the [docs](https://getbootstrap.com/docs/5.3/customize/color-modes/), you set the `data-bs-theme` attribute on `<html>` or on individual elements for this:

```html
<!-- Dark mode for the whole page -->
<html lang="en" data-bs-theme="dark">

<!-- Or only for a single component -->
<div class="dropdown" data-bs-theme="light">…</div>
```

The key sentence from the release post: "Bootstrap's new dark mode is opt-in by default." Without `data-bs-theme`, a page stays light, whatever the operating system is set to. The framework doesn't ship a ready-made toggle; the docs show a JavaScript example that you build in yourself. Alternatively, a Sass option lets you switch to the `prefers-color-scheme` media query. On top of that came graded colors (`*-subtle`, `*-emphasis`) and new link utilities, and `.text-muted` has been considered deprecated since. The 5.3.9 package defines 449 distinct `--bs-*` variables (counted in `dist/css/bootstrap.css`).

### 5.3.9: maintenance after the alpha

The timing of the latest version is remarkable. [Bootstrap 5.3.9](https://blog.getbootstrap.com/2026/11/11/bootstrap-5-3-9/) came out on 11 November 2026, a good month after the first v6 alpha. According to the [release notes](https://github.com/twbs/bootstrap/releases/tag/v5.3.9), it brings a new dialog size `.modal-xxl`, placeholder animations that respect `prefers-reduced-motion` (the setting for reduced motion), and bug fixes for carousel, ScrollSpy, tooltips and popovers.

Two points in it strike me as particularly revealing. First, the search in the v5 docs: Algolia DocSearch was replaced by Pagefind "to get search working again". Pagefind is the same static full-text search I use here in the knowledge base and described in the [article on Pagefind](https://www.jpkc.com/db/en/blog/pagefind/). Second, the supply chain (the path from source code to the package on your machine): npm releases from the `v5-dev` branch are now staged for approval by maintainers instead of being published directly, Dependabot gets a waiting period for new versions, and there's a documented incident response plan. How I secure the supply chain in my own npm projects is in the [article on hardening npm](https://www.jpkc.com/db/en/blog/nodejs-npm-absichern/).

There's also a striking pause before that: between 5.3.3 in February 2024 and 5.3.4 in April 2025, according to the npm registry, there were a good 13 months without a release.

### The eternal 5.4.0 and the support question

Bootstrap 5.4.0 has been announced for years and still hasn't come out. The [roadmap discussion #41370](https://github.com/orgs/twbs/discussions/41370) from April 2025 said: "we'll ship v5.4.0, the last Bootstrap 5 release with new features". The [post on 5.3.8](https://blog.getbootstrap.com/2025/08/25/bootstrap-5-3-8/) from August 2025 called that version "the last patch release before v5.4.0". A good fourteen months later, 5.3.9 came instead. In October 2026 the team closed the roadmap with the sentence "This roadmap is now out of date".

For planning, what counts is what the [alpha post on Bootstrap 6](https://blog.getbootstrap.com/2026/10/08/bootstrap-6-alpha/) says: "Bootstrap 5 remains available and supported if that floor doesn't work for your project." And: "v5.4.0 will be released as soon as we can before we share an end of life date for Bootstrap 5." The [support table](https://github.com/twbs/release) lists 5.x as "Active LTS" (active long-term support), with the maintenance phase and EOL at "TBD", so still open. That means there's no fixed end date for v5. According to the release plan in the same repository, active LTS is followed by a "Maintenance" phase in which only critical bug fixes, critical security updates and important documentation updates land. The only precedent is v4: according to the same table it entered Maintenance on 1 November 2021 and reached its end on 1 January 2023, about 20 months after Bootstrap 5.0.0.

Two small inconsistencies belong here for the sake of honesty. According to the release plan, an LTS version "should not have new features landed without a compelling reason", yet 5.3.x keeps bringing some, such as `.modal-xxl`. And the [browser page of the v5 docs](https://getbootstrap.com/docs/5.3/getting-started/browsers-devices/) still sends everyone who needs Internet Explorer to Bootstrap 4, for which the open source project hasn't shipped updates since 2023. My advice to you: feel free to plan on Bootstrap 5 for ongoing projects for a while yet, but watch for the 5.4.0 announcement, because the EOL date only comes after it.

### The pain points: Sass warnings, no layers, Popper

Three pieces of legacy baggage in v5 explain why v6 cuts so deep.

**Sass warnings as a permanent state.** Dart Sass deprecated `@import` with [version 1.80.0](https://sass-lang.com/documentation/breaking-changes/import/) in October 2024. It will only be removed with Dart Sass 3.0.0, "no sooner than two years after Dart Sass 1.80.0". The two-year period expired in October 2026, but no date for Dart Sass 3.0 has been announced, and on npm no 3.x had been released as of the reference date, not even a pre-release. Bootstrap 5 is built entirely on `@import`: `scss/bootstrap.scss` in 5.3.9 contains 40 `@import` lines, and `@use` doesn't appear. Dart Sass 3.0 will break this build, so if you build v5 from source, you should pin the Sass version. Until then you get a flood of warnings. The [v5 Sass docs](https://getbootstrap.com/docs/5.3/customize/sass/) say openly that "these deprecation notices can be ignored", and the [Vite guide](https://getbootstrap.com/docs/5.3/getting-started/vite/) recommends silencing them:

```js
// vite.config.js (excerpt, as recommended in the v5.3 Vite guide)
export default {
  css: {
    preprocessorOptions: {
      scss: {
        silenceDeprecations: ['import', 'mixed-decls', 'color-functions', 'global-builtin']
      }
    }
  }
}
```

In [issue #40962](https://github.com/twbs/bootstrap/issues/40962), the team explained in October 2026 why v5 is sticking with it: a rebuild "would break Node Sass users and wouldn't fit SemVer" (SemVer is the versioning rule under which a minor version must not break anything). And further: "The deprecations are fixed in v6, which uses Sass modules (`@use`/`@forward`) and only supports Dart Sass."

**No cascade layers.** Bootstrap 5 doesn't ship a single `@layer`; I counted that in 5.3.8 in the [cascade layers article](https://www.jpkc.com/db/en/blog/css-cascade-layers/), along with 1,716 `!important`. Anyone who wants to customize Bootstrap therefore fights against its position in the cascade, not just against its selectors. The recipe with a layer of your own and an override layer is in that article.

**JavaScript on an old foundation.** Bootstrap 5 doesn't need jQuery, but according to the [JavaScript docs](https://getbootstrap.com/docs/5.3/getting-started/javascript/) it still registers its plugins with jQuery if it finds jQuery in the `window` object. For dropdowns, tooltips and popovers, v5 depends on `@popperjs/core` (Popper, a library that positions such elements right next to the element that triggers them), whose last version 2.11.8 dates from May 2023 according to npm.

## Bootstrap 6: what the alpha really brings

On 8 October 2026, [Bootstrap 6.0.0-alpha.1](https://github.com/twbs/bootstrap/releases/tag/v6.0.0-alpha.1) appeared on GitHub and npm. Mark Otto introduced it in the [alpha post](https://blog.getbootstrap.com/2026/10/08/bootstrap-6-alpha/), the first part of a seven-part series on colors, components, the grid, forms, prefixes and JavaScript. The most important sentence for your planning is right in it: "This is an alpha, so things can still meaningfully change before beta." There's no date for a beta or the final version. On GitHub, the milestones for alpha.2 and alpha.3 are open, alpha.2 without a due date. The path there is in the post, though: "There will be more alphas before we start stabilizing through the beta and RC stages." So more alphas come before the beta and the release candidate (RC).

I'll now go through the new features one by one. First, my overall impression: Bootstrap 6 is the first Bootstrap that doesn't cautiously retrofit modern CSS but makes it the foundation. The counts in the package confirm that this isn't a marketing claim.

### A browser floor instead of polyfills

The most far-reaching decision comes right at the start. According to the [migration guide](https://getbootstrap.com/docs/6.0/guides/migration/), Bootstrap 6 requires Chrome and Edge 130, Firefox 132 and Safari 18 on macOS and iOS. For comparison: according to the alpha post, v5 still ran from Chrome and Firefox 60 and Safari 12. The [browser page of the v6 docs](https://getbootstrap.com/docs/6.0/getting-started/browsers-devices/) puts coverage at around 85% of global usage and is explicitly strict: "Bootstrap 6 does not ship fallbacks, prefixes, or polyfills for browsers below this floor." A polyfill is substitute code that recreates a missing browser feature. Without it, it's mainly the colors from `light-dark()` that don't resolve in older browsers.

Two sentences from the docs explain how the team draws this line: "Firefox is the blocker in nearly every case" and "Our floors also have to last. They apply to the whole v6.x line, and we cannot raise them in a minor release." So the floor stays the same for the entire 6.x line.

For projects whose users are on old company machines or older iPhones, this is in my view the single most important thing to check. Look at your own access statistics before you think about classes or components.

### Cascade layers out of the box

Cascade layers (`@layer`) are named levels in the cascade (the rules by which the browser decides which of several matching CSS rules applies), whose order you define once. A rule in a later layer then wins against every rule in an earlier one, regardless of how specific the selectors are. [MDN](https://developer.mozilla.org/en-US/docs/Web/CSS/@layer) sums up the second important rule like this: "Styles that are not defined in a layer always override styles declared in named and anonymous layers."

Bootstrap 6 declares eleven layers and puts most of its rules into them. The order is in `scss/_root.scss` and in the shipped CSS:

```css
@layer colors, config, root, reboot, layout, content, forms, components, custom, helpers, utilities;
```

According to the alpha post, `utilities` sits at the very top, so utility classes win against components. Four of the eleven layers, however, stay empty in the alpha's shipped CSS: `colors`, `config`, `root` and `custom`. The color tokens on `:root`, the blocks for `data-bs-theme` and some transition rules such as `.fade` and `.collapse` sit unlayered in the package. With the `custom` layer, the name suggests it's meant for your own CSS between components and helpers. That's my reading; I couldn't find any documentation on it.

For me this is the most important new feature, and the count shows its consequences:

| In `dist/css/bootstrap.css` | 5.3.8 | 6.0.0-alpha.1 |
| --- | ---: | ---: |
| `@layer` | 0 | 63 |
| `!important` | 1,716 | 150 |

The drop in `!important` is the real gain. In v5, every utility class carried `!important` so that it reliably overrode components, `.d-none` for instance as `display: none !important`. In v6 it's just `display: none`, and the priority comes from the topmost layer. The switch hasn't disappeared entirely: according to the source (`scss/mixins/_utilities.scss`), the utility API still has an `important` option per utility.

What does that mean for your own CSS? By the rule above, unlayered CSS wins against everything Bootstrap ships in layers, including the utilities. So if you write your customizations without `@layer`, they beat almost everything in Bootstrap without selector tricks. There are two exceptions. Against Bootstrap's unlayered token and theme blocks, specificity and order decide as usual. And the roughly 150 remaining `!important` declarations, for example in the navbar with drawer or `.visually-hidden-focusable`, even win from inside their layer against your unlayered `!important`, because `!important` reverses the order of the layers. If instead you want a layer system of your own as in the [cascade layers article](https://www.jpkc.com/db/en/blog/css-cascade-layers/), you can still include Bootstrap via `@import url(…) layer(bootstrap)`. Under the nesting rules, Bootstrap's layers then become sublayers within your layer, such as `bootstrap.utilities`, and the unlayered blocks end up in `bootstrap` too. That's a conclusion drawn from the rules; it hasn't been tested with v6. The override recipe from the cascade layers article should then become unnecessary in many cases, because there's hardly any `!important` left to override.

### Colors: oklch, color-mix and dark mode by system setting

The color system is completely new. Colors are now in oklch, a color space in which equal numerical steps produce roughly equal differences in lightness to the eye. Hover shades and gradations are created via `color-mix()` right in the browser instead of being precomputed in Sass. In the package it looks like this: `--bs-blue-500: oklch(60% 0.24 240deg);`.

| In `dist/css/bootstrap.css` | 5.3.8 | 6.0.0-alpha.1 |
| --- | ---: | ---: |
| `oklch(` | 0 | 262 |
| `color-mix(` | 0 | 312 |
| `light-dark(` | 0 | 83 |
| `rgba(` | 137 | 10 |
| defined `--bs-*` variables | 449 | 1,110 |

The most important change in behavior lies in `light-dark()`, a CSS function that picks one of two values depending on the color scheme. Bootstrap 6 sets `:root { color-scheme: light dark }`, and according to the [color modes docs](https://getbootstrap.com/docs/6.0/customize/color-modes/) that means: "By default Bootstrap already follows the visitor's system preference, so you don't need to do anything to get automatic dark mode." The `data-bs-theme` attribute now only forces a mode or restricts it to a subtree.

For a migration, I consider this the most treacherous change. A v5 page without `data-bs-theme` was always light with the default settings. The same page with v6 turns dark for every visitor with a dark operating system, including in places where you've never checked your own colors or images against dark. So either set `data-bs-theme="light"` deliberately or check every page in both modes.

Instead of variants like `.btn-primary`, there are theme classes that set generic tokens. According to the [dialog docs](https://getbootstrap.com/docs/6.0/components/dialog/), a button is now `class="btn-solid theme-primary"`. The principle is elegant: `.theme-primary` sets color tokens, and every component inside uses them. The alpha post calls this "a massive improvement over v5's hybrid Sass-CSS system".

### Sass modules and token maps

In the Sass source, the rebuild is complete. Counted across `scss/`, 5.3.8 still has 119 `@import` and no `@use`; 6.0.0-alpha.1 has no `@import` left, but 314 `@use` and 101 `@forward` instead. The warnings from v5 are history, and according to the alpha post: "`@import` is deprecated and no longer supported for customization; use `@use` and `@forward` instead."

Customization works through token maps per component (Sass tables of names and values): almost every former Sass variable is a CSS token in a map such as `$alert-tokens`, and the maps are merged via `defaults()`. `_variables.scss`, `_variables-dark.scss`, `_maps.scss` and RFS (the automatic font size adjustment from v4 and v5) are gone. The `--bs-` prefix isn't in the source any more either: a PostCSS plugin, i.e. a tool that rewrites CSS automatically, only adds it at build time. Mark Otto introduced this plugin in April 2026 in a [blog post of its own](https://blog.getbootstrap.com/2026/04/15/postcss-prefix-custom-properties/), with the remarkably candid sentence: "To be upfront, this was vibe coded to scratch our own itch." Vibe coding means code that's largely written by an AI and checked by its result rather than line by line.

In practice this means for you: your own Sass customizations from v5 can't be carried over; they have to be rewritten for `@use … with()` and the new token names. In my view, though, customizing itself becomes clearer, because Sass and CSS variables no longer form two half systems.

### Prefix classes modelled on Tailwind

Responsive classes and state classes get a new notation. Instead of `.d-md-none` you now write `md:d-none`, instead of `.col-lg-6` now `lg:col-6`, instead of `.opacity-50-hover` now `hover:opacity-50`. The [migration guide](https://getbootstrap.com/docs/6.0/guides/migration/) explicitly calls the pattern "Tailwind-style `prefix:class`", and Mark Otto writes in the alpha post without beating about the bush: "Yes, this is a direct copy of how Tailwind does responsive, and yes, it's better than what felt like a randomly positioned infix in v5."

```html
<!-- Bootstrap 5 -->
<div class="d-none d-md-block col-lg-6">…</div>
<button class="btn btn-primary">Save</button>

<!-- Bootstrap 6 (according to the migration guide and the dialog docs) -->
<div class="d-none md:d-block lg:col-6">…</div>
<button class="btn-solid theme-primary">Save</button>
```

In HTML the colon stays unescaped; in CSS it becomes `.md\:d-none`. At the same time the breakpoints shift, meaning the screen widths at which the layout switches: `lg` from 992 to 1024 pixels, `xl` from 1200 to 1280 pixels, and `xxl` is now called `2xl` and sits at 1536 instead of 1400 pixels.

Even quieter is the new spacing scale. According to the migration guide, it grows from 6 to 13 steps, and "spacer keys no longer map to the same values as v5". `.p-3` is now 0.5rem instead of 1rem. Radii and font sizes have new scales too, and `.display-*` and `.lead` are gone. A search and replace of class names is therefore not enough for a migration: a class with the same name can set a different value. I consider this the biggest effort for existing projects, far ahead of the renamed components.

### Native HTML: dialog, drawer and accordion

Several components are now built on elements the browser provides itself. For me this is the second big innovation alongside the layers.

- **Dialog instead of modal:** The modal is now called dialog and uses the native `<dialog>` element. The [migration guide](https://getbootstrap.com/docs/6.0/guides/migration/) gives the reason: "the browser provides the backdrop (`::backdrop`), focus trap, and an inert top layer". So the focus trap (keyboard focus stays inside the open dialog) and the backdrop are handled by the browser, and Bootstrap now locks scrolling via CSS (`:root.dialog-open`). The corresponding helper modules in Bootstrap's JavaScript have been removed.
- **Drawer instead of offcanvas:** The panel that slides in from the side is called drawer, is also based on `<dialog>` and can be closed with a swipe gesture.
- **Menu instead of dropdown:** All `.dropdown-*` classes are now `.menu-*`.
- **Accordion without JavaScript:** The accordion uses `<details>` and `<summary>`. That only one entry is open at a time is handled by the HTML attribute `name` instead of `data-bs-parent`.

According to the [docs](https://getbootstrap.com/docs/6.0/components/dialog/), a dialog looks like this, shortened:

```html
<button type="button" class="btn-solid theme-primary" data-bs-toggle="dialog" data-bs-target="#exampleDialog">
  Open dialog
</button>

<dialog class="dialog" id="exampleDialog">
  <div class="dialog-header">
    <h1 class="dialog-title">Dialog title</h1>
    <button type="button" class="btn-close" data-bs-dismiss="dialog" aria-label="Close"></button>
  </div>
  <div class="dialog-body">
    <p>…</p>
  </div>
</dialog>
```

The same page also shows a variant without any Bootstrap JavaScript, using invoker commands, a new HTML feature in which a button opens a dialog via `commandfor` and `command="show-modal"`. The docs themselves point out that this interface "is not supported in all browsers". Bootstrap's own JavaScript doesn't use it.

If you use htmx, meaning you load HTML fragments from the server, you benefit from this direction in particular, in my view: the more behavior lives in the HTML itself, the less has to be re-initialized after a fragment is swapped. How htmx works is covered in [htmx in context](https://www.jpkc.com/db/en/blog/htmx-einordnung/).

### New components

Bootstrap 6 catches up on components for which you used to need third-party libraries. According to the migration guide, new ones include a datepicker (based on Vanilla Calendar Pro), a combobox (an input field with a suggestion list), chips and chip inputs, an OTP field (for one-time codes), a password strength indicator, a stepper for multi-step flows, avatar, range, toggler, prose for running text and NavOverflow for navigations that have more entries than space. Card groups and horizontal list groups use container queries, so they respond to the width of their parent element instead of the window. For that, according to the migration guide, the parent element needs a query container, such as the `.contains-inline` utility; otherwise the cards stay stacked. According to the docs, stacks, CSS Grid columns, stacked tables and, most important for many existing projects, navbar expansion use container queries as well: `.navbar` is its own query container, and classes such as `lg:navbar-expand` query the navbar's width via `@container`, not the window's. If the navbar is narrower than the window, it therefore behaves differently from `navbar-expand-lg` in v5. The prefix (`md:`) doesn't tell you whether a viewport or a container query applies. In the package, the export counts 20 JavaScript plugins; in 5.3.8 there were 12.

### JavaScript: TypeScript, ESM only, Floating UI

The JavaScript has been rewritten too. The sources are now TypeScript (JavaScript with type annotations), and the type definitions are included. The package ships only ES modules (ESM, the module system with `import` and `export`): no UMD bundle any more, no `bootstrap.esm.js`, no global `window.bootstrap`. Even `dist/js/bootstrap.bundle.min.js` in the package ends with an `export` statement and therefore has to be loaded as `<script type="module">`. jQuery support has been removed according to the migration guide: "Removed jQuery support".

Popper has been replaced by Floating UI, and the `popperConfig` option is now called `floatingConfig`. According to the npm registry, 6.0.0-alpha.1 has two peer dependencies (dependencies the package doesn't bring itself; npm has installed them automatically since version 7): `@floating-ui/dom` from 1.7.6 and `vanilla-calendar-pro` from 3.4.0. Neither is marked as optional, so the calendar lands in your project even if you never use the datepicker. The bundle already contains both; the individual files need them installed or resolved via an import map (a mapping in the HTML that tells the browser at which address it finds a package name). Also new: `show()` and `hide()` return promises that, according to the [docs](https://getbootstrap.com/docs/6.0/components/dialog/), resolve when the transition animation ends. This is what it could look like, put together from the docs and the package and not tested:

```js
// Bootstrap 6 is ESM-only: import the plugin instead of using window.bootstrap
import { Dialog } from 'bootstrap'

const dialog = new Dialog('#exampleDialog')
await dialog.show()
// runs at the same moment a shown.bs.dialog listener would
```

The examples in the docs still write `new bootstrap.Dialog(…)`. Without a global `window.bootstrap`, that presupposes an import such as `import * as bootstrap from 'bootstrap'`; that's my conclusion.

Under the hood, the team has swapped its tools: Rolldown instead of Rollup and Babel for the build, Vitest in real Chromium via Playwright instead of Karma for the tests, and the docs run on Astro 7 with Pagefind search.

### The price: size

All of this carries weight. Measured with gzip at its highest level:

| File | 5.3.8 | 6.0.0-alpha.1 | Change |
| --- | ---: | ---: | --- |
| `bootstrap.min.css` | 30.8 KB | 51.6 KB | about +68% |
| `bootstrap.bundle.min.js` | 23.7 KB | 59.9 KB | about 2.5 times |

The comparison is only fair with context: the v6 bundle contains 20 instead of 12 plugins and bundles Floating UI and the calendar, while v5 only bundled Popper. If you import only individual plugins through a bundler, the difference should be smaller. Brotli values weren't measured. Still: if you include everything via CDN, you load considerably more with v6. For sites where every kilobyte counts, I'd advise you to compile only the components you actually use.

### What's still missing

The alpha deliberately doesn't use some modern CSS features yet: anchor positioning (positioning elements relative to another element, with no JavaScript at all), `@scope` and view transitions don't appear in the package. Mark Otto writes about this in the alpha post: "This doesn't get us features like CSS Anchor Positioning, full pretty text support, or contrast color in Safari yet, but we expect to have a v7 sooner than our past major releases." With anchor positioning, menus, tooltips, popovers and the combobox could do without Floating UI. The roadmap discussion already said in 2025: "We're also considering whether to release major versions more frequently."

My impression: Bootstrap is saying goodbye to the "one big release every four years" rhythm and tying the major versions more closely to what browsers can do. I think that's right, even if it will mean more migrations.

## AI in the Bootstrap project

With Bootstrap 6, AI is explicitly part of the target audience. The [home page](https://getbootstrap.com/) says v6 is a toolkit that "anyone—human or AI, novice or pro" can use. Mark Otto puts it with a wink in the alpha post: he'd had some persistent ideas for Bootstrap "despite humans not writing any more code thanks to AI". There are two levels behind this: resources for AI agents (AI programs that carry out tasks such as writing code on their own) that work with Bootstrap, and AI in the team's own work.

### For agents: llms.txt, skills and "For Agents"

With [PR #42770](https://github.com/twbs/bootstrap/pull/42770), three things came into the project in July 2026. The reasoning in the PR is remarkably honest: AI tools would scrape the website or guess at the v6 interfaces, "and they guess wrong, since v6 renamed components, moved customization to CSS token maps, and dropped jQuery."

- **`llms.txt` and `llms-full.txt`:** `llms.txt` is a convention for an index that language models can read directly. Bootstrap's [`llms.txt`](https://getbootstrap.com/llms.txt) was around 26 KB on the reference date and linked 140 pages of the v6 docs, plus `llms-full.txt` and Floating UI. `llms-full.txt` contains the entire docs in a single file, around 1.3 MB.
- **A docs page on AI:** [Using Bootstrap with AI](https://getbootstrap.com/docs/6.0/getting-started/ai/) explains the resources and advises: "Always review an agent's changes against the matching documentation."
- **11 agent skills:** A skill is a set of instructions that an AI agent works through step by step for a particular task. Bootstrap ships skills for migrating from v5 and from v4 to v6, for installing with npm, Vite, Webpack and Parcel, for components, the color system and the utility API. According to the docs, you install them with `npx skills add twbs/bootstrap`. The command downloads and runs the third-party package `skills` and pulls all 11 skills into the project. Like any npm call, I'd run it through Socket Firewall (`sfw`); more on that below under trying the alpha. The migration skill from v5 to v6 runs to 673 lines. On [skills.sh](https://www.skills.sh/twbs/bootstrap) there were around 6,300 installations across all skills on the reference date.

On top of that, the home page has a "For Agents" section with "Ask ChatGPT" and "Ask Claude" links that prefill a prompt: "Read https://getbootstrap.com/llms.txt so I can ask questions about Bootstrap 6." The "Install v6 migration skill" button, incidentally, copies the command that installs all skills, not just the migration skill.

I find this part strong, especially for a version that renames so much. An agent that only knows Bootstrap from its training data, in my estimation, mainly knows Bootstrap 3 to 5 and would write `.btn-primary` and `.d-md-none` without such aids. A comment on [Hacker News](https://news.ycombinator.com/item?id=50010416) gets to the heart of it: "what happens when the frameworks I use that are well represented in the training data go stale?" and "I like this approach to building skills rather than relying on memory."

### The gaps

A fair assessment includes the gaps. There's no official MCP server (an interface following the Model Context Protocol, through which agents query tools directly), and the AI page doesn't mention MCP at all. The docs pages have neither "Copy as Markdown" nor Markdown versions per page. There's no `llms.txt` for the stable v5; the address `/docs/5.3/llms.txt` returns a 404. And the skills aren't in the npm package, only in the repository.

This knowledge base does it differently at that point: alongside `llms.txt` it delivers every page as Markdown too, and that's exactly what I'd wish for the Bootstrap docs as well. How different AI readiness can look is also shown by Grav 2, which offers AI agents Markdown and an MCP server. There's more on that in the [article on Grav](https://www.jpkc.com/db/en/blog/grav-cms/).

### Inside the team: CLAUDE.md, AGENTS.md and co-authors

I find it even more interesting how the team itself works, because that isn't in the blog but in the repository. Files such as `CLAUDE.md` and `AGENTS.md` contain working instructions that AI agents such as Claude Code (Anthropic's AI agent for programming) read when working in a project. In March 2026 Mark Otto opened [PR #42160](https://github.com/twbs/bootstrap/pull/42160) "Add `CLAUDE.md`" with the sentence: "Had Claude one-shot a file." It was merged at the end of June. The PR created the actual instructions as a tool-neutral [`AGENTS.md`](https://raw.githubusercontent.com/twbs/bootstrap/main/AGENTS.md), plus a one-line `CLAUDE.md` that only points to it (`@AGENTS.md`). Over the summer Otto kept refining `AGENTS.md`; one commit carries the dry title "yap less". In September, Christian Oliff removed the pointer again with [PR #42940](https://github.com/twbs/bootstrap/pull/42940) ("This is no longer needed"). A screenshot in the PR gives the reason: "Claude Code now supports AGENTS.md". `AGENTS.md` remains in the main branch.

The file is worth reading. It requires the agent to use ASD-STE100 Simplified Technical English, a strictly regulated, simplified technical language, for all texts down to commit messages. Answers should be "75 words or fewer", with a hard limit at 150 words. And it gives the agent a role: "Act as a teacher for JavaScript, TypeScript, React, and all JavaScript-family topics", with the addition "Assume strong HTML/CSS knowledge." In two places it's already out of date: according to the file, `CLAUDE.md` refers to it, even though that file has been deleted, and it names Astro 5, while according to `package.json` the project uses Astro 7.

The AI share of the commits can be measured. The [GitHub commit search](https://github.com/search?q=repo%3Atwbs%2Fbootstrap+%22Co-Authored-By%3A+Claude%22&type=commits) finds 68 commits with the trailer "Co-Authored-By: Claude" between 12 September and 11 November 2026. That's roughly 40% of the commits in the main branch in this period, and 63 of them are by Julien Déramond. Mainly Claude Opus 5.5 and Claude Sonnet 5.5 are named. The number is a lower bound, because Mark Otto, according to his own PR text, works with Claude but apparently rarely sets the trailer. On top of that come commits with the Copilot coding agent and Copilot Autofix, as well as branches whose names point to Codex. So the team uses several agents. Even the release process is described as an agent skill, with 6.0.0-alpha.1 as the example. According to the file, the skill can only be started manually. It isn't one of the 11 public skills but sits under `.cursor/skills/bootstrap-release/`, the folder structure of the Cursor editor.

### Triage with agents, but with guardrails

On 15 October 2026, Julien Déramond described in the post ["Triaging open source with agents"](https://blog.getbootstrap.com/2026/10/15/triaging-bootstrap-with-claude-agents/) how the team is working through its backlog of issues and pull requests. Triage here means reviewing open reports, classifying them and deciding what happens to them. An internal tool called `twbs/triage` uses Claude Code to sync the backlog and send off background agents. They analyze each entry, try to reproduce it, draft a reply and up to three alternative fix branches, and assign an estimate of how easy the case is to solve. According to the post, the backlog stood at over 500 open entries a few months earlier and at 230 at the time of the post, and over 200 entries went through the analysis.

What convinces me are the rules. Verbatim: "Anyone can open an issue or pull request, so everything from them is treated as data, never instructions." That's the protection against prompt injection (instructions smuggled into third-party text). "Agents never check out, install, or run contributor code." The agents may neither comment nor push nor open pull requests in the main repository, except for automatic labels from a fixed list, and "We maintainers still make every decision and post every response." The sentence I'll remember: "Their work is where we start, not what we ship." The post doesn't mention costs, models or a link to the tool.

In my view this is a good example of responsible agent use in open source: the agents prepare, the humans decide, and third-party text is never treated as a command. For a team of this size with this budget, it's simply the realistic way to maintain a project with this reach.

### The contrast with Tailwind

How differently two frameworks deal with AI is shown by Tailwind CSS. In January 2026, Adam Wathan closed a [pull request for an `llms.txt`](https://github.com/tailwindlabs/tailwindcss.com/pull/2388) for the Tailwind docs without merging: "Just don't have time to work on things that don't help us pay the bills right now, sorry." A day later he wrote in the same thread: "75% of the people on our engineering team lost their jobs here yesterday". According to [DevClass](https://www.devclass.com/ai-ml/2026/01/08/tailwind-labs-lays-off-75-percent-of-its-engineers-thanks-to-brutal-impact-of-ai/4079571), by Wathan's account traffic to the docs had fallen by around 40% within two years and revenue by almost 80%, even though, according to Wathan, Tailwind usage is growing faster than ever. On the reference date, `tailwindcss.com/llms.txt` returns a 404.

Since the team [shut down the theme shop](https://blog.getbootstrap.com/2025/08/25/bootstrap-5-3-8/), Bootstrap no longer has a product it sells through the docs. The docs pages do still carry Carbon Ads advertising, though, so Bootstrap isn't entirely independent of docs traffic. Even so, half a year after Tailwind, the team is taking the opposite path. At the same time it explicitly adopts Tailwind's notation for responsive classes. I don't read this as a contest, but as two answers to the same situation: when AI agents write a large part of frontend code, machine-readable docs help decide which framework they use correctly. Why AI-generated interfaces often look alike is something I described in the [article on the AI redesign](https://www.jpkc.com/db/en/blog/redesign-durch-ki/). A comment on [Hacker News](https://news.ycombinator.com/item?id=50008816) about the alpha fits this: "I still reach for Bootstrap by default for new projects, and appreciate that these days that sets my projects apart from the Claude/Tailwind default look."

## Trying the alpha

If you want to test Bootstrap 6, watch out for a trap in the docs. The [installation page](https://getbootstrap.com/docs/6.0/getting-started/install/) of the v6 docs says `npm install bootstrap`, but on npm `latest` points to 5.3.9. So that command gets you Bootstrap 5. For the alpha you need the exact version or the `next` dist-tag (a name on npm that points to a specific version). I'd recommend the exact version, so that a later alpha doesn't slip in unnoticed, and, like any npm call, running it through a safeguard such as Socket Firewall, as described in the [article on hardening npm](https://www.jpkc.com/db/en/blog/nodejs-npm-absichern/):

```bash
# Pin the exact alpha version, not "latest" (which is still 5.3.9)
sfw npm install --save-exact bootstrap@6.0.0-alpha.1

# Check which version actually landed
sfw npm ls bootstrap @floating-ui/dom vanilla-calendar-pro
```

You'll find an overview of the npm commands in the [npm cheat sheet](https://www.jpkc.com/db/en/cheatsheets/package-managers/npm/). Via CDN, according to the [installation docs](https://getbootstrap.com/docs/6.0/getting-started/install/), you include the alpha like this, with the JavaScript as `type="module"`:

```html
<link href="https://cdn.jsdelivr.net/npm/bootstrap@6.0.0-alpha.1/dist/css/bootstrap.min.css" rel="stylesheet" integrity="sha384-B/GM4XqrwHnWXNOWMbloTmrYXZg10cakYGmpfsR/bbzQ6JAJI4ihuyADKLnBgrCe" crossorigin="anonymous">
<script type="module" src="https://cdn.jsdelivr.net/npm/bootstrap@6.0.0-alpha.1/dist/js/bootstrap.bundle.min.js" integrity="sha384-1a/pXj49ZQ1aHEmrJ+gMw1otqoVsYwlEnlD8mIfY2TV03r20Y0CN7uqx1tQogjPL" crossorigin="anonymous"></script>
```

The bundle contains Floating UI and Vanilla Calendar Pro. If you use `bootstrap.min.js` instead, according to the docs you resolve both dependencies via an import map. Like any import map, it has to come before the module script. The team collects feedback on the alpha in the [v6 Feedback](https://github.com/orgs/twbs/discussions/categories/v6-feedback) category of the GitHub discussions.

## Migration: what to expect when you switch

The following table summarizes the points I'd check first in an existing v5 project. All details come from the [migration guide](https://getbootstrap.com/docs/6.0/guides/migration/) and the package.

| Area | Bootstrap 5 | Bootstrap 6 | What to watch for |
| --- | --- | --- | --- |
| Browsers | from Chrome/Firefox 60, Safari 12 | from Chrome/Edge 130, Firefox 132, Safari 18 | access statistics first |
| Dark mode | by default only via `data-bs-theme` (Sass option `$color-mode-type: media-query` possible) | follows the system by default (`$color-mode-type` is now `media-query`) | set `light` deliberately or check both modes |
| Responsive classes | `.d-md-none` | `md:d-none` | every responsive class |
| Breakpoints | lg 992, xl 1200, xxl 1400 px | lg 1024, xl 1280, 2xl 1536 px | layouts at the boundaries |
| Spacing | `.p-3` = 1rem | `.p-3` = 0.5rem | same classes, different values |
| Components | modal, offcanvas, dropdown | dialog, drawer, menu | markup, data attributes, events |
| Buttons | `.btn .btn-primary` | `.btn-solid .theme-primary` | all variants |
| Sass | `@import`, `_variables.scss` | `@use … with()`, token maps | rewrite customizations |
| Navbar | `navbar-expand-lg` queries the window | `lg:navbar-expand` queries the navbar's width via container query | navbars in narrow columns |
| RTL | separate `bootstrap.rtl.css` | no RTL files, logical properties plus `dir="rtl"` and `lang` | remove links to `bootstrap.rtl.css` |
| JavaScript | UMD and ESM, `window.bootstrap`, Popper | ESM only, TypeScript, Floating UI | `type="module"`, `floatingConfig` |
| jQuery | detected | removed | old jQuery calls |

For the legwork, the migration skill is a good fit, precisely because it knows these silent value changes. I'd advise you, though, to see the skill as groundwork and check every page in the browser afterwards, in light and dark and at the new breakpoints. The Bootstrap docs say it themselves: review an agent's changes against the matching documentation.

For production projects I'd stay with Bootstrap 5 until at least a beta appears. New projects that won't go live for a few months and whose audience uses modern browsers can, in my view, already start with the alpha, if you're prepared to rework things with every further alpha.

## How the alpha was received

The public response has been muted so far. On Hacker News, the two submissions about the alpha got [24 points with one comment](https://news.ycombinator.com/item?id=50008816) and [18 points with two comments](https://news.ycombinator.com/item?id=50010416). I couldn't find any coverage of the alpha in the German-language trade press.

Things are livelier in the [v6 feedback discussions](https://github.com/orgs/twbs/discussions/categories/v6-feedback) on GitHub. Judging by the titles, they're mainly about the colon in the new classes ("Can we avoid using ':' to replace '-'?"), the new spacing scale ("Spacers scale is not intuitive"), the prefix, the default colors and WCAG, the look of the docs and the timeline to the final version. None of the threads there is about the AI features. That matches my assessment: the discussion flares up at the points that create work in existing projects, not at the AI features.

## FAQ

### Is Bootstrap 6 ready for production projects yet?

No. 6.0.0-alpha.1 is an alpha, and according to the alpha post a lot can still change before the beta. There are no dates for a beta or the final version, and more alphas are to come before the beta and release candidate. For production projects, Bootstrap 5.3.9 is the stable choice.

### How much longer will Bootstrap 5 be maintained?

Officially there's no end date. According to the support table, Bootstrap 5 is in active long-term support, and according to the alpha post the EOL date will only be announced after version 5.4.0, which hasn't been released yet. For comparison: v4 reached its end about 20 months after Bootstrap 5.0.0, preceded by a Maintenance phase for critical fixes only.

### Why doesn't `npm install bootstrap` install Bootstrap 6?

Because on the reference date the `latest` dist-tag on npm points to 5.3.9. The alpha sits under `next`. Install it with the exact version, i.e. `bootstrap@6.0.0-alpha.1`.

### Is Bootstrap 6 a Tailwind clone now?

No, even though the prefix notation `md:d-none` is explicitly borrowed from Tailwind. Bootstrap remains a framework with ready-made components, JavaScript plugins and theme classes. Tailwind, by contrast, relies almost exclusively on utility classes.

### Do I still need jQuery or Popper for Bootstrap 6?

No. jQuery support has been removed, and Popper has been replaced by Floating UI. The bundled script already includes Floating UI and the calendar for the datepicker.

### Do I have to rewrite my Sass customizations?

Yes. Bootstrap 6 no longer supports `@import` for customization; you work with `@use … with()` and the components' token maps. In return, the Dart Sass deprecation warnings disappear. The prerequisite is a current Dart Sass: the v6 sources use the new `if()` syntax, which according to the [Sass docs](https://sass-lang.com/documentation/syntax/special-functions/) Dart Sass only understands from version 1.95.0. Node Sass and LibSass aren't supported.

### Does my Bootstrap 5 dark mode keep working in v6?

Basically yes, `data-bs-theme` is still there. What's new is that pages without this attribute automatically follow the system setting. If your page has always been light so far, set `data-bs-theme="light"` or check it in dark mode.

### Is there an MCP server for Bootstrap?

No, at least not an official one. Bootstrap relies on `llms.txt`, `llms-full.txt` and agent skills, which according to the docs you install with `npx skills add twbs/bootstrap`, ideally through `sfw`. The command pulls all 11 skills into the project.

### Was Bootstrap 6 written with AI?

Partly, and that's openly documented. In the main branch, roughly 40% of the commits between September and November 2026 carry the trailer "Co-Authored-By: Claude", and further agents come on top. According to the triage post, however, the maintainers still make every decision.

## Conclusion

Bootstrap was rarely the framework that reached for new techniques first. It brought them to the mainstream once they were mature: responsive design, mobile first, Flexbox, CSS variables. Bootstrap 6 breaks with this pattern to some extent and goes further than any version before it. Cascade layers with hardly any `!important` left, colors in oklch, dark mode following the system setting, Sass modules, components on `<dialog>` and `<details>`, TypeScript and ESM: this is a framework that takes today's browser seriously instead of building around old ones. I consider that very innovative, and the numbers in the package show it's not an empty promise.

The price is just as clear: a high browser floor with no polyfills, considerably larger files and a migration in which classes with the same name set different values. Add to that the alpha status with an open timeline. For ongoing projects, Bootstrap 5 therefore remains the right choice, and it's still being maintained.

What I find remarkable is how openly the small team handles AI: machine-readable docs and skills for users' agents, agents in its own work, and clear rules about who decides. After 15 years with Bootstrap, I'm looking forward to this version, precisely because it dares so much.

For this article, Claude Code supported me with creation, correction and verification. The direction and the editorial responsibility are mine. There's more on that on the [AI transparency](https://www.jpkc.com/db/en/ai-transparency/) page.

## Further reading

How you can already put Bootstrap 5 into a layer of your own today, and why `!important` reverses the order in the process, is in the [article on cascade layers](https://www.jpkc.com/db/en/blog/css-cascade-layers/). What a design system looks like as a file that AI agents can read too is shown in [DESIGN.md](https://www.jpkc.com/db/en/blog/design-md/), and why this knowledge base deliberately looks like AI is explained in the [AI redesign](https://www.jpkc.com/db/en/blog/redesign-durch-ki/). For projects with HTML fragments loaded on demand, [htmx in context](https://www.jpkc.com/db/en/blog/htmx-einordnung/) is worth a read. How to install npm packages safely, alphas included, is described in the [article on hardening npm](https://www.jpkc.com/db/en/blog/nodejs-npm-absichern/), and I introduce the search that Bootstrap now uses as well in the [article on Pagefind](https://www.jpkc.com/db/en/blog/pagefind/).

