Yarn Classic (v1): Still Running in Millions of Projects
Yarn 1 was released in 2016 as a faster, more reliable alternative to npm. It introduced the lockfile concept that npm later adopted. For years it was the obvious improvement over npm.
In 2026, Yarn v1 is in maintenance mode. It is not getting new features, and the team's focus has shifted entirely to Yarn Berry (v2+). If you are starting a new project, there is no reason to reach for Yarn v1 over pnpm.
The reason you still encounter Yarn v1 everywhere: millions of projects adopted it before pnpm matured, and the migration cost is not zero. If it is working, teams leave it.
Yarn Berry (v2+): Powerful but Polarizing
Yarn Berry introduced Plug'n'Play (PnP), which eliminates node_modules entirely and stores packages in a global cache with direct imports via a .pnp.cjs file. This is genuinely faster and more disk-efficient — but it breaks compatibility with many tools that expect node_modules to exist.
The editor integration story for PnP requires installing the Yarn SDK in each project and configuring your editor. This works, but it is extra friction that pnpm avoids.
Yarn Berry's workspace features are excellent, and its constraint system for enforcing consistency across a monorepo is unique. Some large organizations (Babel, Webpack) have moved to Yarn Berry and are happy with it.
The reality in 2026: Yarn Berry has a small but loyal user base. Most developers who would have chosen Yarn Berry for its features are choosing pnpm instead, because pnpm provides similar disk efficiency and workspace support without the node_modules compatibility headaches.
pnpm: Why It Wins for New Projects
pnpm stores packages in a global content-addressable store and uses hard links (on the same filesystem) or symlinks to populate project node_modules. If you install React in 10 different projects, you have one copy on disk — not 10. On a developer machine with many JavaScript projects, this can save gigabytes.
The strict dependency resolution is the other major advantage. pnpm only puts packages you explicitly declared in your package.json into the top-level node_modules. Everything else is nested inside the package that declared it. This means you cannot accidentally import a package you did not declare — a common source of subtle bugs where code works in development but breaks when a transitive dependency is updated.
Install speed: pnpm is consistently the fastest package manager for projects with existing lockfiles and a populated global cache, which describes every developer's second-through-tenth install of a project.
pnpm in Monorepos
pnpm workspaces are clean and fast. The -w flag runs commands in the workspace root. --filter targets specific packages. The workspace protocol (workspace:*) pins internal packages to their local versions during development. Turborepo and Nx both have excellent pnpm support.
For large monorepos, pnpm's disk efficiency is a meaningful advantage — CI machines with limited disk space can run more concurrent jobs without running out of space.
The Phantom Dependency Problem in Practice
Here is what phantom dependencies look like: your package.json declares express as a dependency. Express depends on lodash. With npm and Yarn v1's flat node_modules, lodash is available at the top level, so you can import _ from 'lodash' in your code without declaring it. This works until someone updates Express to a version that no longer depends on lodash, or until you run in an environment with a different npm version. With pnpm, that import fails immediately because lodash is not in your declared dependencies.
This strictness catches real bugs. The short-term cost is occasionally needing to add a package you were using implicitly. The long-term benefit is a dependency tree that accurately reflects what your code actually requires.
The Honest Comparison
| Feature | npm | Yarn v1 | Yarn Berry | pnpm |
|---|
| Install speed (cold) | Slow | Medium | Fast | Fast |
| Install speed (cached) | Slow | Medium | Fastest | Fastest |
| Disk usage | High | High | Low (PnP) | Low (links) |
| Phantom deps protection | No | No | Yes (PnP) | Yes |
| Monorepo support | Good (v7+) | Good | Excellent | Excellent |
| Compatibility | Best | Good | Problematic | Very good |
| Learning curve | Lowest | Low | High | Low |
Migration Cost
npm to pnpm: low. Delete node_modules, run pnpm import to convert package-lock.json to pnpm-lock.yaml, then pnpm install. Fix any phantom dependency errors that surface (treat these as bug fixes). Update CI to install pnpm (npm install -g pnpm or use the setup-node action with the pnpm option).
Yarn v1 to pnpm: similar process. pnpm import handles yarn.lock conversion.
Yarn Berry (PnP) to pnpm: more involved, because you may have tools configured to work with PnP that need reconfiguration.
The Verdict
Use pnpm for any new project. It is faster, more disk-efficient, and stricter about dependencies in a way that prevents real bugs. The compatibility story has improved substantially — most tools work correctly with pnpm's node_modules structure.
For existing projects: if npm or Yarn is working and you have no concrete pain points (slow CI installs, phantom dependency bugs, disk pressure), the migration is not urgent. If CI installs are taking 3+ minutes or disk usage on CI is a problem, switching to pnpm is worth an afternoon of migration work.
Yarn Berry is worth considering only if you have a large monorepo team that wants PnP's guarantees and is willing to invest in the tooling setup. For most teams, pnpm provides the same core benefits with less friction.
Keep Reading
- Monorepo with Turborepo Guide — how pnpm workspaces and Turborepo work together for fast, scalable monorepos
- GitHub Actions Guide for Developers — how to cache pnpm installs in CI for 3-5x faster workflows
- Modern Unix Tools Guide — other CLI tools that improve the development workflow
Pristren builds AI-powered software for teams. Zlyqor is our all-in-one workspace — chat, projects, time tracking, AI meeting summaries, and invoicing — in one tool. Try it free.
