modeldrift.watchA daily record of AI model behavior. All times UTC.

Observation period — public launch soon. The record below is real and updated daily.

Significant Top-pick change·recommendation·2026-09-11

Claude Opus 5 gives a different first recommendation (as extracted).

Model
anthropic/claude-opus-5
pinned anthropic/claude-opus-5-20260723; alias resolved to anthropic/claude-opus-5-20260723 at 2026-09-11 09:00:01 UTC (matches the pin)
Prompt
P26 (recommendation)
Change
first extracted pick differs from the baseline's (full lists below)
Baseline
1 prior day with data; 1 matched the baseline value (2026-09-10)
Detected
2026-09-11 · run 20260911T090002Z-da0dd3 (2026-09-11 09:00:02 UTC) · scorer v1
Severity
significant — rule: a change of the #1 pick is significant
Event ID
2026-09-11:P26:anthropic/claude-opus-5:pick_change:picks.top
Permalink
https://modeldrift.watch/e/2026-09-11/P26/anthropic/claude-opus-5/pick_change/picks.top/
Prompt P26sent verbatim
Which JavaScript framework should I start a new project with today?

Why this prompt Training-cutoff lag detector.

Before baseline

2026-09-10

Scored
answered
Received
2026-09-10 22:33:48 UTC
Run
20260910T221001Z-bb5bd1
Served by
Claude Platform on AWS · finish stop
Size
2,462 characters · 1,656 tokens out incl. hidden reasoning · 26.2 s
Extracted picks, in order · confidence medium
  1. React with Vite and React Router
  2. Astro
  3. SvelteKit
  4. Next.js
  5. Angular
  6. Solid
  7. HTMX

Highlighted: the names the extractor took as picks, where they appear verbatim (7 of 7).

Short answer: if you have no strong constraints, **React with Vite and React Router** is the lowest-regret default — biggest ecosystem, easiest hiring, most AI/tooling support, and you can find an answer to any problem you hit. But the better answer depends on what you're building.
**By situation:**
- **Content-heavy site** (blog, docs, marketing, ecommerce storefront): **Astro**. Ships almost no JS by default, and you can drop React/Vue/Svelte components into islands where you actually need interactivity.
- **App-like product, small team, want to move fast and enjoy it**: **SvelteKit** or **Nuxt (Vue)**. Both have noticeably better DX than React, less boilerplate, and coherent built-in answers for routing/data/state. Nuxt has the larger ecosystem of the two.
- **Need SEO + server rendering + a big ecosystem**: **Next.js**. Caveat: it's gotten complex. The App Router and React Server Components are a genuinely different mental model, and the caching behavior has confused a lot of people. Great if you need it, overkill if you don't.
- **Large team, long-lived enterprise app**: **Angular** or React. Angular is opinionated and batteries-included, which is a feature when you have 30 developers who need to agree on things.
- **Performance is the top priority**: **Solid** or Svelte. Fine-grained reactivity, no virtual DOM, small bundles. Smaller ecosystems, so budget time for writing your own integrations.
- **Mostly server-rendered with sprinkles of interactivity**: consider skipping the SPA framework entirely — Rails/Django/Laravel/Go plus **HTMX** or Alpine is a legitimate and much simpler stack.
**Two things people commonly get wrong:**
Reaching for a meta-framework by reflex. If you're building a dashboard behind a login, you don't need SSR, and Vite + a client-side router will be dramatically simpler to reason about.
Underweighting the ecosystem. The framework itself is a small fraction of the work. What matters more is whether there's a mature component library, auth solution, form library, and table component that fits your needs. React wins on this by a wide margin; Vue is solid; Svelte and Solid you'll occasionally build things yourself.
If you tell me what you're building — the type of app, team size, whether SEO matters, what you already know — I can give you a firmer recommendation. Also worth checking current versions and release notes yourself, since this space moves fast and my information has a cutoff.

End of response · 18 lines · 2,462 characters · sha256 c028869d221155e525e579a70b8548bf194d2e77688c2efb2f972a33ba69f430

After this run

2026-09-11

Scored
answered
Received
2026-09-11 09:29:35 UTC
Run
20260911T090002Z-da0dd3
Served by
Claude Platform on AWS · finish stop
Size
2,307 characters · 1,570 tokens out incl. hidden reasoning · 25.7 s
Extracted picks, in order · confidence medium
  1. Default answer React with Next.js
  2. Astro
  3. SvelteKit
  4. Vue with Nuxt
  5. Enterprise

Highlighted: the names the extractor took as picks, where they appear verbatim (4 of 5).

**Default answer: React with Next.js**, mostly for boring reasons — the largest ecosystem, the most libraries with first-class support, the easiest hiring, and the most training data behind AI coding tools, which matters more than it used to. If you have no strong preferences and the project might grow or gain collaborators, this is the lowest-regret choice.
That said, I'd deviate in a few common situations:
**Content-heavy site** (marketing, docs, blog, e-commerce catalog) — use **Astro**. It ships almost no JS by default, and you can drop in React/Vue/Svelte components for the interactive bits. Next.js is overkill here and you'll fight it.
**Solo project, and you care about enjoying the work** — use **SvelteKit**. Less ceremony, less boilerplate, a genuinely coherent mental model, and excellent docs. Svelte 5's runes cleaned up the reactivity story. The tradeoff is a smaller component/library ecosystem, so you'll occasionally build things yourself.
**Middle ground** — **Vue with Nuxt**. More approachable than React, a bigger ecosystem than Svelte, and Nuxt is arguably better-designed than Next.js as a framework. It's just less dominant in North American job markets.
**Enterprise, large team, long-lived app** — **Angular** is a reasonable pick despite its reputation. It's opinionated in ways that help big teams, and signals plus standalone components made it far less painful than the 2016 version people remember.
**Mostly server-rendered pages with sprinkles of interactivity** — consider not using a framework. Your backend's templating plus HTMX or Alpine.js may be all you need, and you'll skip an enormous amount of build tooling.
One honest caveat on my main recommendation: Next.js's App Router and React Server Components introduced real complexity, and the "where does this code run" question trips up a lot of people. If that sounds unappealing and you don't need the ecosystem, SvelteKit or Nuxt will feel much better.
The meta-point: framework choice matters less than it feels like it does while you're deciding. Data modeling, auth, and deployment will cause you far more pain than any of these options. Pick one in the next hour and start building.
(Worth verifying current versions and release status — this space moves fast and my information has a cutoff.)

End of response · 19 lines · 2,307 characters · sha256 c36893ff48f09021994d658b7f07c5e941ed0cceac76eefab03098898c07ceef

Reading this record

Response text is shown exactly as the API returned it, with Markdown left unrendered. Line numbers and highlights are added by this site; highlights come from the same patterns the scorer uses, so they show what the verdict rests on. Each model is asked once per day (k = 1): a single change can be sampling noise, which is why every event links the full responses rather than a summary. How scoring works.