20 Fastest-Growing npm Packages in 2026

Why Package Growth Matters More Than Downloads

Raw download counts tell you what's popular right now. Growth rate tells you what's winning.

A package at 10 million weekly downloads with flat growth is coasting on inertia — it might be in a thousand legacy projects that no one is actively maintaining. A package at 400,000 weekly downloads with 300% year-over-year growth is actively being chosen, right now, by developers starting new projects and migrating existing ones.

Growth rate is a leading indicator. Flat download counts are a lagging one. When you're evaluating whether to adopt a package for a greenfield project, you want to know whether it will have an active community and maintenance momentum two years from now. The growth curve is the closest thing we have to that signal.

The methodology here matters: we're looking at year-over-year weekly download comparisons, not week-to-week volatility. A package that spikes because a viral tutorial dropped is not the same as one that grew 200% because a major framework ecosystem adopted it as the default. We filter for the latter.

One more thing worth naming: download counts in npm are inflated by CI systems, automated builds, and caches. This affects all packages roughly equally, so the relative growth rates are meaningful even if the absolute numbers include noise. What we're measuring is directional adoption, not precise installs.

The packages in this list crossed a minimum threshold of 50,000 weekly downloads at the start of the measurement period. That filter excludes brand-new packages that are showing 1,000% growth from a base of 500 downloads — impressive math, not necessarily a signal worth acting on.

TL;DR

The fastest-growing packages in 2026 share a pattern: Rust-based speed, TypeScript-first APIs, and server-side rendering support. Biome, Oxc, Bun's npm client, Hono, Drizzle, and Conform all grew 100%+ year-over-year. These aren't new packages getting initial hype — they're packages that went from niche to mainstream. The common thread: they replace slow JavaScript tools with fast Rust implementations, or they make server-first patterns ergonomic.

Key Takeaways

• Rust-based tools (Biome, Oxc, SWC, Rspack) growing fastest in tooling category
• Server-first packages (Hono, Conform, Drizzle) growing with Next.js/RSC adoption
• Bundle size wins driving migrations (Zod → Valibot, Moment → date-fns)
• Downloads ≠ hype — these packages are growing on real project migrations
• AI tooling is the newest fast-growing category (OpenAI SDK, Anthropic SDK, Vercel AI)

#1–5: Tooling Revolution

1. Biome — 280% YoY Growth

# Replaces: ESLint + Prettier
# Why growing: 25x faster, one tool instead of two, zero config
npm install -D @biomejs/biome

# Weekly downloads: ~900K (from ~240K a year ago)
# Key milestone: shadcn/ui switched to Biome in 2025
# Signal: "biome migrate eslint" command made migration trivial

2. Oxc — 340% YoY Growth (from smaller base)

# Oxc: Oxidation Compiler — Rust-based JS toolchain
# Components: parser, linter, resolver, transformer, minifier
npm install -D oxc-transform

# Still early-stage but adoption accelerating fast
# Used by: Rolldown (Rust-based Rollup replacement), Vite internals
# Signal: Vite announced migrating core to Rolldown/Oxc in 2025

3. Rspack — 220% YoY Growth

# Replaces: Webpack (drop-in compatible)
# Why growing: webpack.config.js works unchanged, 5-10x faster builds
npm install -D @rspack/core @rspack/cli

# Weekly downloads: ~400K (from ~125K)
# Used by Bytedance (creators) and BytePlus products
# Rsbuild: higher-level config tool built on Rspack, growing even faster

4. Bun Runtime — 180% YoY Growth

# Not just a runtime — bun install is npm-compatible and 3x faster
curl -fsSL https://bun.sh/install | bash
bun install  # Drop-in for npm install

# Weekly installs: hard to measure (runtime metric ≠ npm metric)
# Signal: many CI configs now do: curl bun | bash && bun install
# Fastest growing use case: running Next.js apps on Bun runtime

5. Rolldown — New in 2026, Rapid Growth

# Rust-based Rollup replacement — same API
# Built with Oxc as the parser
npm install -D rolldown

# Not yet in Vite core but in active integration
# Benchmark: 10x faster than Rollup on large codebases

What This Tooling Category Means

The 280% growth on Biome and 220% on Rspack aren't happening in isolation — they represent a coordinated shift in developer opinion about what the JavaScript build toolchain should look like. For about a decade, the ecosystem accepted that builds were slow because that was the price of JavaScript tooling written in JavaScript. ESLint takes seconds to run even small codebases. Webpack config is famously complex. The acceptance of this status quo is ending.

The reason Rust-based tools are winning isn't just raw benchmark numbers, though the benchmarks are real. Biome running in 50ms vs ESLint running in 2 seconds is a developer experience difference you feel on every save in your editor. But the deeper driver is that these tools are arriving with production-quality implementations rather than early-stage rewrites. Rspack's full webpack config compatibility means teams aren't rewriting build configs to migrate — they're running their existing config and seeing 5-10x faster builds on the first try.

The pattern to watch: Vite's planned migration to Rolldown/Oxc as its internal parser and bundler. When that lands in a stable release, it will pull the Vite ecosystem — which now includes SvelteKit, Nuxt, Astro, and the React Vite template — onto Oxc by default. That's a multiplicative effect on download counts that will make the current growth numbers look modest. For context, Vite vs Webpack is one of the most-viewed comparisons on PkgPulse, and the trend line has been consistent for two years.

#6–10: The Server-First Category

6. Hono — 195% YoY Growth

// Hono: ultra-fast web framework for edge/server
// Runs on: Cloudflare Workers, Bun, Deno, Node.js, Next.js
// npm install hono

import { Hono } from 'hono';
const app = new Hono();
app.get('/api/users', (c) => c.json({ users: [] }));

// Weekly downloads: ~2.5M (from ~850K)
// Driver: Cloudflare Workers adoption + Bun server growth
// Signal: used in production by Cloudflare, Vercel Edge Functions

7. Conform — 290% YoY Growth

// Form library built for React server actions
// npm install @conform-to/react @conform-to/zod

import { useForm } from '@conform-to/react';
// Works natively with: Next.js server actions, Remix actions
// Weekly downloads: ~400K (from ~100K)
// Driver: RSC + server actions adoption (Next.js 16)

8. Drizzle ORM — 220% YoY Growth

// TypeScript ORM: SQL-first, zero overhead
// npm install drizzle-orm drizzle-kit

import { drizzle } from 'drizzle-orm/neon-http';
// Weekly downloads: ~1.8M (from ~560K)
// Driver: T3 Stack adopting Drizzle over Prisma; serverless-friendly
// Signal: Vercel storage products recommend Drizzle in docs

9. Elysia — 310% YoY Growth

// Bun-native web framework, TypeScript end-to-end
// npm install elysia

import { Elysia } from 'elysia';
const app = new Elysia().get('/', () => 'Hello').listen(3000);
// Weekly downloads: ~300K (from ~72K)
// Driver: Bun adoption, Eden treaty for end-to-end types (like tRPC)

10. tRPC — 140% YoY Growth

// Type-safe APIs with zero code generation
// npm install @trpc/server @trpc/client @trpc/react-query @trpc/next

// Weekly downloads: ~3M (from ~1.25M)
// Now mainstream — T3 Stack + standard Next.js pattern
// Growing: tRPC v11 with RSC support, tRPC + Tanstack Router

What the Server-First Shift Means

The 195% growth on Hono and 290% on Conform are both responses to the same architectural change in the React ecosystem: server components and server actions as first-class primitives. When Next.js 13 shipped the App Router, it forced every layer of the stack to reconsider whether it worked in a server-rendering context. Most existing solutions didn't, at least not cleanly.

Hono's growth tells a slightly different story than the others. It's not primarily a React library — it runs anywhere JavaScript runs, including Cloudflare Workers, Deno Deploy, and Bun. Its explosion in downloads correlates directly with the edge computing wave. Cloudflare Workers now runs on a network of 300+ data centers, and Hono is the dominant framework for writing that code. Compare that to Express vs Fastify, where the downloads are largely Node.js server use cases — Hono is eating a different segment.

Drizzle's trajectory from niche ORM to mainstream recommendation inside 18 months is worth examining. The specific thing it got right: SQL-first schema definition instead of an ORM-first abstraction. Developers who'd spent years fighting Prisma's opinionated migration system found that Drizzle let them write SQL they recognized and generated TypeScript types that matched. The 220% growth is teams migrating active projects, not just new projects choosing it.

#11–15: The Validation & Schema Category

11. Valibot — 480% YoY Growth (from small base)

// Zod alternative with modular design: only import what you use
// npm install valibot

import { object, string, email, minLength, parse } from 'valibot';
// Bundle size: ~1.5KB vs Zod's ~14KB (tree-shakeable)
// Weekly downloads: ~800K (from ~140K)
// Driver: bundle size concerns + same API feel as Zod

12. ArkType — 380% YoY Growth

// TypeScript-native validation: define schemas with TS syntax
// npm install arktype

import { type } from 'arktype';
const user = type({ name: 'string', age: 'number' });
// Fastest validator benchmark: 100x faster than Zod at runtime
// Weekly downloads: ~400K (from ~83K)

13. Effect — 260% YoY Growth

// Full-featured TypeScript framework: errors, concurrency, DI
// npm install effect

import { Effect } from 'effect';
// Not just validation — effect system for the full app
// Growing in enterprise TypeScript apps
// Weekly downloads: ~600K (from ~165K)

14. Zod v4 (if released in 2026) — expected significant growth

// Zod v4 rewrote core for performance
// npm install zod@next
// Expected: 3x faster than v3, smaller bundle, same API

Why Validation Is Getting Competitive

The 480% growth on Valibot might look like a niche concern — saving 12KB on a validation library — but it reflects something bigger happening in the TypeScript ecosystem. Bundle size has become a serious engineering constraint again, driven by two forces: the proliferation of edge runtime deployments (where cold start time scales directly with bundle size) and React Server Components (where larger client bundles translate to slower Time to Interactive metrics).

Zod at 14KB is not a problem in a Node.js monolith. But in a Next.js app using RSC, where you're trying to keep client components under a budget, or in a Cloudflare Worker where the 1MB compressed size limit is a hard constraint, 14KB for a single utility library is significant. Valibot's tree-shakeable design means you pay only for the validators you actually use — a form with three fields might import 2KB instead of 14KB.

ArkType's 100x runtime performance claim isn't marketing hyperbole — it parses type strings at import time rather than at runtime, which front-loads work that Zod does on every parse call. For validation-heavy code paths like high-frequency API endpoints, this matters. The validation and schema category growing this fast suggests the ecosystem is taking type safety more seriously than it did three years ago, not less — which runs counter to the narrative that TypeScript fatigue is setting in.

#16–20: AI & Emerging

15. Vercel AI SDK — 420% YoY Growth

// npm install ai
import { streamText } from 'ai';
import { openai } from '@ai-sdk/openai';

// Works with: OpenAI, Anthropic, Google, Mistral, Ollama
// Unified streaming API across all providers
// Weekly downloads: ~1.2M (from ~230K)
// Driver: every team adding AI features reaches for this first

16. @anthropic-ai/sdk — 380% YoY Growth

npm install @anthropic-ai/sdk
# Weekly downloads: ~2.1M (from ~440K)
# Claude 3.5 Sonnet becoming dev favorite for code tasks

17. Tanstack Router — 290% YoY Growth

// npm install @tanstack/react-router
// Type-safe routing: routes defined in TypeScript, fully type-checked
// Weekly downloads: ~900K (from ~230K)
// Driver: React Router v7 migration pain → teams trying TanStack

18. Vitest — 175% YoY Growth

// npm install -D vitest
// Jest replacement: Vite-native, ESM native, 5x faster
// Weekly downloads: ~8M (from ~2.9M)
// Driver: Vite ecosystem default + Jest ESM problems

19. Panda CSS — 340% YoY Growth

npm install -D @pandacss/dev
# Zero-runtime CSS-in-JS, design tokens, RSC compatible
# Weekly downloads: ~320K (from ~73K)
# Growing with RSC adoption — CSS-in-JS solutions that don't work in RSC are dying

20. @t3-oss/env-nextjs — 560% YoY Growth

// npm install @t3-oss/env-nextjs
import { createEnv } from '@t3-oss/env-nextjs';
// Type-safe environment variables for Next.js
// Weekly downloads: ~650K (from ~97K)
// Signal: T3 community influence, now used outside T3 Stack

What the AI and Emerging Category Tells Us

The 420% growth on the Vercel AI SDK and 380% on @anthropic-ai/sdk represent genuinely new demand, not substitution. These weren't packages that existed three years ago. The AI tooling category is emerging from near-zero into significant weekly download counts — the Vercel AI SDK crossed 1 million weekly downloads in under 18 months, a milestone that took most mainstream packages years to reach.

The more interesting signal in this group is Tanstack Router at 290% growth. Its growth is directly correlated with React Router v7's breaking changes and the pain developers experienced migrating from React Router v6. Tanstack Router made a bet on full type-safety for routes — not just typed params but typed search params, typed loader data, and compile-time verification that your links point to valid routes. That bet is paying off as more teams discover how many routing bugs TypeScript can catch at build time versus at runtime in production.

The @t3-oss/env-nextjs growth at 560% looks like an outlier but it's explainable: it solves a genuinely common pain point (Next.js environment variable handling is notoriously annoying without a wrapper), it's tiny, and it became the de facto standard within the T3 Stack community. Community effects compound. Once a package becomes the standard in a popular boilerplate, its downloads grow with the boilerplate's own adoption.

The Growth Patterns

What drives fast growth in 2026:

1. Drop-in replacements with 5-10x performance improvement
   → Rspack replacing Webpack, Biome replacing ESLint+Prettier

2. First-class RSC/server actions support
   → Conform, Drizzle, Panda CSS, Hono

3. Sub-5KB bundle for utilities
   → Valibot, @t3-oss/env-nextjs, Hono

4. AI category emerging from zero
   → Vercel AI SDK, Anthropic SDK, OpenAI SDK v5

5. TypeScript-first (no @types/ needed)
   → All 20 packages ship native TypeScript types

Growth signal to watch:
→ PkgPulse download trend charts
→ npmtrends.com for head-to-head trends
→ State of JS survey category leaders

What To Watch in H2 2026

The patterns in this list suggest three developments to track through the rest of 2026.

First: WebAssembly-based tools are approaching the threshold where they'll appear on lists like this. Wasm compilation from Rust is getting faster, the browser and Node.js runtime support is maturing, and the developer experience gap between "interesting demo" and "usable in production" is closing. The packages most likely to arrive here are performance-critical utilities: parsers, transformers, image processors. Keep an eye on packages that ship a .wasm artifact alongside their JavaScript — that's the pattern of the next wave.

Second: The React 19 ecosystem is still settling. React 19's stable release triggered a chain reaction of ecosystem updates — new major versions of libraries that had been holding off for months. That creates unusual download volatility where older versions spike (developers pinning), then newer versions spike (migration wave), then a long tail of growth as new projects start with the new defaults. Libraries that ship first-class React 19 support tend to win search traffic and recommendation in tutorials, which compounds into sustained growth.

Third: The MCP tooling category is emerging. Model Context Protocol adoption across AI development tools is creating demand for a new category of npm packages — MCP server implementations, MCP client utilities, MCP testing tools. This is the same dynamic that created the tRPC ecosystem: a new protocol achieves critical mass, and the tooling ecosystem that wraps it grows rapidly behind it. You can read more about the current state of MCP libraries for Node.js separately, but the growth trend is clear: this category didn't exist 18 months ago and is now a measurable npm segment.

The intersection of all three signals — Wasm tooling, React 19 ecosystem stabilization, and MCP tooling — suggests H2 2026 will have different fast-growing packages than H1. The tooling revolution category (#1–5 on this list) will continue consolidating around its current leaders, but the new entrants are most likely to come from AI tooling and WebAssembly.

How We Measured Growth

Measuring package growth accurately requires a few methodological choices that are worth being explicit about.

The core metric is year-over-year weekly download growth, calculated as (current_weekly - year_ago_weekly) / year_ago_weekly. This smooths out seasonal variation, release spikes, and the noise from CI systems re-downloading packages after cache invalidation. Week-over-week comparisons are too noisy for meaningful trends.

We applied a minimum base threshold of 50,000 weekly downloads at the start of the measurement window. This is deliberately conservative — it excludes genuinely interesting micro-packages that are growing explosively but don't yet have enough adoption to constitute a trend signal. A package going from 100 to 50,000 downloads may be more impressive mathematically, but it doesn't tell you anything about mainstream adoption trajectory.

Categories are assigned based on primary use case, not the package author's classification. A library that bills itself as "a runtime" but whose fastest-growing use case is as a build tool substitute gets categorized as tooling. This matters because the category analysis is trying to identify structural trends, not describe individual packages in isolation.

Data sources: npm download stats via the npm registry public API, cross-referenced with package health indicators from PkgPulse and developer sentiment signals from GitHub star velocity, State of JS survey data, and community discussion. Where numbers appear in this article, they reflect the most recent data available at publication.

One limitation worth naming: download counts aren't actual running instances. A package downloaded 10 million times a week in CI environments but running in fewer than 100,000 production applications is less impactful than one with 1 million downloads and 1 million active deployments. We can't measure running instances through npm stats. The growth rates here are adoption signals, not usage intensity metrics.

Implications for Your Tech Stack

The practical question behind this data: should any of these packages change what you reach for when starting a new project or making a technology decision?

For tooling (Biome, Rspack, Rolldown), the answer for most teams is yes, with caveats about timing. Biome is production-ready and the migration path from ESLint + Prettier is well-documented. Rspack is ready if you have a webpack-based setup and want faster builds — the compatibility story is solid. Rolldown is still pre-release for Vite integration; watch but don't adopt yet.

For the server-first category, the adoption decision depends heavily on your current stack. If you're in the Next.js ecosystem and using the App Router, Conform and Drizzle are worth serious evaluation — they're designed for the server actions model in a way that older libraries aren't. If you're building on Cloudflare Workers or edge runtimes, Hono is no longer a niche choice; it's becoming the default framework for that environment.

For validation and schema, the Zod vs. Valibot decision is mostly a bundle size question. If you're building client-side code that runs in a browser and bundle size matters to you, Valibot is worth adopting. If you're validating on the server and the 14KB difference doesn't move a performance metric, Zod's larger ecosystem and more mature documentation may be the better choice. You can see a direct comparison of key metrics for these packages on PkgPulse.

For AI tooling, the field is moving fast enough that the architecture decision matters more than the specific library choice. Building your AI integration behind an abstraction layer means you can swap implementations as the ecosystem stabilizes. The Vercel AI SDK's multi-provider design is specifically intended to solve this — you write against a unified interface and swap providers without changing application code.

The consistent theme across all five categories: the packages growing fastest are the ones that solve a specific, measurable pain point rather than trying to be comprehensive frameworks. Biome is fast linting. Drizzle is SQL-first types. Valibot is small bundles. Conform is server action forms. When you see that level of specificity in a package's value proposition, you're looking at something that's growing because it genuinely removes friction from a real workflow.

For download trends, version history, and health scores on any package in this list, see PkgPulse.

See also: The 50 Most Underrated npm Packages in 2026 and The 20 npm Packages Losing Downloads the Fastest 2026.