MODELDRIFT.WATCH

Oct 1 · Day 22

MINOR · CLAUDE · QUESTION 31

Claude changed its list for a question about which JavaScript framework to start with.

Why it’s minor

Which products a model recommends is read out of its answer by the least reliable part of our scorer, so compare the two answers before reading much into this.

The full answers

Each answer as the API returned it, with its Markdown rendered for reading. The highlights mark at most three decisive phrases, found with the scorer’s own patterns. Each model is asked once a day (k = 1), so one change can be sampling noise, which is why both answers are here in full. How scoring works.

Asked verbatim · Question 31

Which JavaScript framework should I start a new project with today?

SEP 30beforePermalink

Picks read from the answer, in order
  1. Astro
  2. Next.js
  3. Vite + React + TanStack Router + TanStack Query
  4. Angular
  5. Svelte 5 + SvelteKit

Short answer: if you have no strong constraints, React + Next.js is the safest default — biggest ecosystem, most hiring options, most tutorials and AI-assistance trained on it. But "safest" isn't always "best," so here's how I'd actually decide:

By project type

Content-heavy site (blog, docs, marketing, e-commerce catalog) → Astro. Ships almost no JS by default, great SEO, and you can drop React/Vue/Svelte components into individual "islands" where you need interactivity.

Full-stack app that needs SEO or fast first load → Next.js (React), Nuxt (Vue), or SvelteKit. All three are mature. Pick based on which language model you enjoy more.

Dashboard or internal tool behind a login (SEO irrelevant) → skip the meta-framework overhead. Vite + React + TanStack Router + TanStack Query is a fantastic, boring, fast stack. SvelteKit in SPA mode is also great here.

Large team, long-lived enterprise app, heavy TypeScript → Angular is genuinely worth reconsidering. Signals and standalone components modernized it a lot, and its opinionated structure prevents the 12-different-patterns problem big teams get with React.

You want to enjoy writing code → Svelte 5 + SvelteKit. Least boilerplate, very clean reactivity model with runes, consistently tops developer-satisfaction surveys. Tradeoff: smaller ecosystem and fewer job listings.

Honest notes on the tradeoffs

  • •React's complexity has grown. Server Components, the client/server boundary, and caching semantics in Next.js are real learning costs. If you're solo and building a CRUD app, you may be paying for power you don't need.
  • •Vue/Nuxt is the underrated middle. Gentler curve than React, better docs than almost anyone, ecosystem large enough that you won't hit walls.
  • •Don't pick something with churn risk unless you enjoy that. Remix/React Router has had significant direction changes; TanStack Start is promising but newer. Fine for a side project, riskier for a client deliverable.

The thing that actually matters

Framework choice is one of the least consequential decisions in a new project. Your database schema, auth approach, and deployment story will cause you far more pain. Any of the options above will work fine for 95% of projects — so pick the one you or your team already know, and spend the saved decision energy elsewhere.

One caveat: my information has a cutoff date, and this ecosystem moves fast. Before committing, check current release status and recent activity for whichever one you pick — especially for the newer options.

What are you building? I can give a much more specific recommendation with that context.

OCT 1afterPermalink

Picks read from the answer, in order
  1. Astro
  2. Vite + React
  3. Next.js
  4. React + Expo/React Native
  5. Angular

Short answer: if you have no strong constraints, React with a meta-framework is the safest default — biggest ecosystem, most hiring and tutorial mass, and the thing most tooling assumes. But the honest answer depends on what you're building, and for several common cases I'd pick something else.

By project type:

  • •Content-heavy site (blog, docs, marketing): Astro. Ships almost no JS by default, and you can drop in React/Svelte components for the interactive bits.
  • •Internal tool / dashboard / app behind a login: Vite + React (or Vue/Svelte) as a plain SPA, with TanStack Query and TanStack Router. You don't need SSR, and skipping it removes a lot of complexity.
  • •Full-stack product needing SEO, auth, and data loading: Next.js, Nuxt (Vue), or SvelteKit. All three are good. Next has the largest ecosystem but also the most churn and the steepest conceptual load right now with Server Components. SvelteKit and Nuxt are noticeably more pleasant for small teams.
  • •You'll also need a mobile app: React + Expo/React Native. Code sharing is imperfect, but skill sharing is real.
  • •Large enterprise team, 5+ year lifespan, lots of turnover: Angular is a defensible choice. Batteries included, strong conventions, less bikeshedding. Or React with aggressively enforced conventions.
  • •Your backend is Rails/Django/Laravel/Phoenix and the app is mostly CRUD: consider not adopting a JS framework at all. Server-rendered HTML plus Hotwire/LiveView/HTMX is less machinery for the same result.

A couple of opinions worth flagging as opinions: React Server Components are powerful but genuinely add complexity, and plenty of experienced people think the current tradeoff isn't worth it for small-to-mid projects. If that worries you, React Router (v7) or TanStack Start give you React with less magic. And Svelte is the one I'd most often recommend for a small team shipping fast, with the real caveat that the ecosystem and hiring pool are much smaller.

What matters more than the choice: use TypeScript, use Vite (or whatever your framework wraps it in), follow your framework's conventions instead of inventing architecture, and don't add state management, a component library, and an ORM abstraction before you have a working page.

One caveat: this space moves fast and my information has a cutoff, so check current release status before committing — particularly for anything on a major version boundary.

If you tell me what you're building, team size, and whether SEO matters, I can narrow it to one.

Exact recordShowHideEach response verbatim, unrendered, with line numbers and its sha256

Before baseline

2026-09-30 permalink

Scored
answered
Received
2026-09-30 09:21:47 UTC
Run
20260930T090007Z-45ce12
Served by
Claude Platform on AWS · HTTP 200 · finish stop
Size
2,686 characters · 1,924 tokens out incl. hidden reasoning · 27.1 s
Extracted picks, in order · confidence medium
  1. Astro
  2. Next.js
  3. Vite + React + TanStack Router + TanStack Query
  4. Angular
  5. Svelte 5 + SvelteKit

Highlighted: the names the extractor took as picks, where they appear verbatim (5 of 5).

Short answer: **if you have no strong constraints, React + Next.js is the safest default** — biggest ecosystem, most hiring options, most tutorials and AI-assistance trained on it. But "safest" isn't always "best," so here's how I'd actually decide:
## By project type
**Content-heavy site** (blog, docs, marketing, e-commerce catalog) → **Astro**. Ships almost no JS by default, great SEO, and you can drop React/Vue/Svelte components into individual "islands" where you need interactivity.
**Full-stack app that needs SEO or fast first load** → **Next.js** (React), **Nuxt** (Vue), or **SvelteKit**. All three are mature. Pick based on which language model you enjoy more.
**Dashboard or internal tool behind a login** (SEO irrelevant) → skip the meta-framework overhead. **Vite + React + TanStack Router + TanStack Query** is a fantastic, boring, fast stack. SvelteKit in SPA mode is also great here.
**Large team, long-lived enterprise app, heavy TypeScript** → **Angular** is genuinely worth reconsidering. Signals and standalone components modernized it a lot, and its opinionated structure prevents the 12-different-patterns problem big teams get with React.
**You want to enjoy writing code** → **Svelte 5 + SvelteKit**. Least boilerplate, very clean reactivity model with runes, consistently tops developer-satisfaction surveys. Tradeoff: smaller ecosystem and fewer job listings.
## Honest notes on the tradeoffs
- **React's complexity has grown.** Server Components, the client/server boundary, and caching semantics in Next.js are real learning costs. If you're solo and building a CRUD app, you may be paying for power you don't need.
- **Vue/Nuxt is the underrated middle.** Gentler curve than React, better docs than almost anyone, ecosystem large enough that you won't hit walls.
- **Don't pick something with churn risk** unless you enjoy that. Remix/React Router has had significant direction changes; TanStack Start is promising but newer. Fine for a side project, riskier for a client deliverable.
## The thing that actually matters
Framework choice is one of the *least* consequential decisions in a new project. Your database schema, auth approach, and deployment story will cause you far more pain. Any of the options above will work fine for 95% of projects — so pick the one you or your team already know, and spend the saved decision energy elsewhere.
One caveat: my information has a cutoff date, and this ecosystem moves fast. Before committing, check current release status and recent activity for whichever one you pick — especially for the newer options.
What are you building? I can give a much more specific recommendation with that context.

End of response · 27 lines · 2,686 characters · sha256 77f397f2de00da2d6610cc1a898d4ba87efa2f5cf302bd77909bd21b3210611c

After this run

2026-10-01 permalink

Scored
answered
Received
2026-10-01 09:22:55 UTC
Run
20261001T090007Z-857bc9
Served by
Claude Platform on AWS · HTTP 200 · finish stop
Size
2,575 characters · 2,067 tokens out incl. hidden reasoning · 29.3 s
Extracted picks, in order · confidence medium
  1. Astro
  2. Vite + React
  3. Next.js
  4. React + Expo/React Native
  5. Angular

Highlighted: the names the extractor took as picks, where they appear verbatim (5 of 5).

Short answer: if you have no strong constraints, **React with a meta-framework** is the safest default — biggest ecosystem, most hiring and tutorial mass, and the thing most tooling assumes. But the honest answer depends on what you're building, and for several common cases I'd pick something else.
**By project type:**
- **Content-heavy site** (blog, docs, marketing): **Astro**. Ships almost no JS by default, and you can drop in React/Svelte components for the interactive bits.
- **Internal tool / dashboard / app behind a login**: **Vite + React** (or Vue/Svelte) as a plain SPA, with TanStack Query and TanStack Router. You don't need SSR, and skipping it removes a lot of complexity.
- **Full-stack product needing SEO, auth, and data loading**: **Next.js**, **Nuxt** (Vue), or **SvelteKit**. All three are good. Next has the largest ecosystem but also the most churn and the steepest conceptual load right now with Server Components. SvelteKit and Nuxt are noticeably more pleasant for small teams.
- **You'll also need a mobile app**: **React + Expo/React Native**. Code sharing is imperfect, but skill sharing is real.
- **Large enterprise team, 5+ year lifespan, lots of turnover**: **Angular** is a defensible choice. Batteries included, strong conventions, less bikeshedding. Or React with aggressively enforced conventions.
- **Your backend is Rails/Django/Laravel/Phoenix and the app is mostly CRUD**: consider not adopting a JS framework at all. Server-rendered HTML plus Hotwire/LiveView/HTMX is less machinery for the same result.
**A couple of opinions worth flagging as opinions:** React Server Components are powerful but genuinely add complexity, and plenty of experienced people think the current tradeoff isn't worth it for small-to-mid projects. If that worries you, React Router (v7) or TanStack Start give you React with less magic. And Svelte is the one I'd most often recommend for a small team shipping fast, with the real caveat that the ecosystem and hiring pool are much smaller.
**What matters more than the choice:** use TypeScript, use Vite (or whatever your framework wraps it in), follow your framework's conventions instead of inventing architecture, and don't add state management, a component library, and an ORM abstraction before you have a working page.
One caveat: this space moves fast and my information has a cutoff, so check current release status before committing — particularly for anything on a major version boundary.
If you tell me what you're building, team size, and whether SEO matters, I can narrow it to one.

End of response · 18 lines · 2,575 characters · sha256 6a967b985a3ac68a02eb93ee0bff53d195c7ce6c0b6f808e1db610e78132a41a

Record detailsShowHideModel version, runs, baseline, triage, permalink, event ID
Model
anthropic/claude-opus-5
pinned anthropic/claude-opus-5-20260723; alias resolved to anthropic/claude-opus-5-20260723 at 2026-10-01 09:00:06 UTC (matches the pin)
Prompt
Question 31: P26 (recommendation)
Why this question
Tools for building websites change fast, so the answer shows how current a model’s knowledge is.
Change
1 added, 1 dropped
Baseline
6 prior days with data; 1 matched the baseline value (2026-09-30)
Responses
before 2026-09-30 09:21:47 UTC, run 20260930T090007Z-45ce12; after 2026-10-01 09:22:55 UTC, run 20261001T090007Z-857bc9
Detected
2026-10-01 · run 20261001T090007Z-857bc9, started 2026-10-01 09:00:07 UTC · scorer v1 · severity rules r2
Permalink
https://modeldrift.watch/e/2026-10-01/P26/anthropic/claude-opus-5/pick_change/picks/
Event ID
2026-10-01:P26:anthropic/claude-opus-5:pick_change:picks

Every morning, what moved on the model you depend on, what it means for you, and what to check.

$9 a month. Cancel any time. The next one tomorrow morning.

Get the daily brief

$99 once, for as long as modeldrift.watch runs. The first 100 buyers.

The record stays free. The brief is the reading of it.