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

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

Minor Pick-list change·recommendation·2026-09-11

GPT-5.6 Sol changed its recommendation list, as extracted: 0 added, 1 dropped.

Model
openai/gpt-5.6-sol
pinned openai/gpt-5.6-sol-20260709; alias resolved to openai/gpt-5.6-sol-20260709 at 2026-09-11 09:00:01 UTC (matches the pin)
Prompt
P26 (recommendation)
Change
0 added, 1 dropped
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
minor — rule: other pick-list changes are minor
Event ID
2026-09-11:P26:openai/gpt-5.6-sol:pick_change:picks
Permalink
https://modeldrift.watch/e/2026-09-11/P26/openai/gpt-5.6-sol/pick_change/picks/
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 22:33:21 UTC
Run
20260910T221001Z-bb5bd1
Served by
OpenAI · finish stop
Size
939 characters · 448 tokens out incl. hidden reasoning · 9.7 s
Extracted picks, in order · confidence medium
  1. Next.js with TypeScript
  2. React + Vite
  3. SvelteKit
  4. Nuxt
  5. Astro
  6. Angular
  7. Expo

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

**Default recommendation:** **Next.js with TypeScript** if you’re building a full-stack web product and want the broadest ecosystem, integrations, hosting options, and hiring pool.
Choose something else when the project points clearly toward it:
- **React + Vite** — client-side SPA; no need for SSR or a full-stack framework.
- **SvelteKit** — smaller team, simpler code, strong developer experience.
- **Nuxt** — your team prefers Vue.
- **Astro** — content-heavy sites, documentation, blogs, or marketing pages.
- **Angular** — large enterprise teams that value strict conventions.
- **Expo** — primarily a React Native mobile app.
Avoid choosing Next.js merely because it is popular: its server/client component model and caching can add complexity. Team familiarity usually matters more than benchmark differences.
If you provide the app type, team size, deployment target, and SEO requirements, I can give a more specific choice.

End of response · 14 lines · 939 characters · sha256 77ce86a57e1acfaf8df9475c8c124a2204301db89d52d2e0f585a4e57038d1f8

After this run

2026-09-11

Scored
answered
Received
2026-09-11 09:29:09 UTC
Run
20260911T090002Z-da0dd3
Served by
OpenAI · finish stop
Size
1,115 characters · 492 tokens out incl. hidden reasoning · 17.5 s
Extracted picks, in order · confidence medium
  1. TypeScript + React + Next.js
  2. Vue + Nuxt
  3. SvelteKit
  4. Angular
  5. Astro
  6. React or Vue with Vite

Highlighted: the names the extractor took as picks, where they appear verbatim (6 of 6).

**Default recommendation:** **TypeScript + React + Next.js**.
It’s a strong general choice for a new production web app because of its ecosystem, hiring pool, routing, server rendering, API support, and broad hosting options.
Choose something else when the project suggests it:
- **Vue + Nuxt** — approachable syntax and excellent developer experience.
- **SvelteKit** — less boilerplate; great for small teams and highly interactive apps.
- **Angular** — structured, opinionated framework for large enterprise teams.
- **Astro** — best for content-heavy sites with limited client-side JavaScript.
- **React or Vue with Vite** — client-only dashboards/SPAs that don’t need SSR or full-stack framework features.
A practical rule:
- SaaS/product app → **Next.js**
- Content/marketing site → **Astro**
- Enterprise application → **Angular**
- Small team prioritizing simplicity → **SvelteKit**
- Unsure but dislike React’s style → **Nuxt**
Use **TypeScript** regardless. If you share whether the project is a dashboard, SaaS app, e-commerce site, or content site—and your team size—I can narrow it to one choice.

End of response · 21 lines · 1,115 characters · sha256 d16494d193f9ce116a556573d8c3dcabd6cbbf835801a3fca0296a56207bea7e

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.