UCP Checker
Product

Pin Your Models: UCP Playground's New Model Menu and 11 New Models

· 14 November 2026 · 6 min read

UCP Playground has a new model menu and eleven new models. The models you pin now sit at the top, every other model sits under its lab, and older models wait one click away. On any run's page, Same task, another model sets up the same task at the same store with another model in one step. Both shipped on 14 November 2026, in Playground 0.24.204 (the models) and 0.24.205 (the menu).

Why it matters: when people come back to Playground they rarely stay on one model. Most try several, usually from more than one lab, on the same store, and keep returning to the same few. The old menu listed every model by lab with nothing to tell them apart, so comparing meant scrolling the same long list every time.

Try it: open Runs at any UCP store, or at one of our test stores, open the Model pill and pin the models you use.

The model menu on Runs: three pinned models at the top, then every model under its lab, A to Z, with "new" on models added in the last 30 days and "default" on Gemini 3.6 Flash.

The model menu

The menu is one list, in this order:

  • Pinned: your models, in the order you pinned them. They are not repeated further down.
  • Every other model, by lab: Alibaba, Anthropic, DeepSeek, Google, Meta, Moonshot AI, OpenAI and xAI, A to Z, each lab's newest model first.
  • Show older models: earlier versions, opened in the same list. They still run and keep their history on the leaderboard.
  • Edit pinned models… at the end.

Two small words help you read it. new marks a model added in the last 30 days; after that it is just another model. default marks the model a run starts on, Gemini 3.6 Flash. Typing a model's name jumps to it wherever it sits, older models included.

Pin your models

Pin a model with the pin on its row; the row itself still picks the model, so you can pin without switching. On a phone or with a keyboard, Edit pinned models… opens a short list of every model with a box beside each. You can pin up to eight.

Edit pinned models: every current model under its lab with a box beside each; three are ticked.

You don't need an account to pin. Signed-out visitors keep their pins in that browser, and the pins come along when they sign in.

Older models stay one click away

Earlier versions, such as Claude Opus 5, GPT-5.5 and Gemini 3.1 Pro now that newer models from their labs are here, sit under Show older models. Nothing about them changes: it still runs, and its runs stay on the leaderboard. They just stop crowding the top of the menu.

Show older models opens them at the end of the same list: Claude Opus 5, Claude Sonnet 5, GPT-5.5, Gemini 3.1 Pro and the rest.

A model is only retired when it can't run any more. If a model you had chosen is retired, Runs starts on the default instead and says so once, in one line, in the Info pane.

Same task, another model

A common next step after a run is to run the same task again with a different model. On a run's page, Same task, another model opens the same menu, without the model you just used.

A finished run at the Flowers test store, with the Same task, another model menu open beside Share.

Pick one and Runs opens at the same store with the same task and that model, ready to send. It is never sent for you: you see exactly what will run, and press Send.

Runs after picking GPT-6.1 Sol: the same task in the message box and GPT-6.1 Sol in the Model pill.

Eleven new models

Lab Added
Alibaba Qwen 3.8 Max
Anthropic Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5
Google Gemini 4 Argon, Gemini 4 Hydrogen
Meta Muse Ember
Moonshot AI Kimi K3.1 Pro
OpenAI GPT-6.1 Sol, GPT-6.1 Terra
xAI Grok 4.7

Each runs as its own model with its own record, so the leaderboard shows "metrics pending" until it has runs. Before they went live, each new model made a tool call and finished a test run at our Flowers test store.

Two models are retired: DeepSeek R1 and o4-mini, because the upstream model provider no longer serves them. Their past runs stay on the leaderboard. That brings Playground to 35 models from 8 labs; the full list is in Choosing models.

How the list stays in order

Every model now has a place in its life cycle: new for its first 30 days, current, older once newer models from its lab take its place, and retired when it can't run any more.

A daily check also asks the upstream model provider whether every model is still served. One that has gone missing two days running, and fails a test call, is hidden from the menu until it comes back. It is not retired automatically. That is what would have caught DeepSeek R1 and o4-mini sooner.

FAQ

How do I pin a model in UCP Playground?

Open the Model pill on Runs and use the pin on a model's row, or choose Edit pinned models… at the end of the menu and tick the ones you want. Pinned models stay at the top, in the order you pinned them.

Do I need an account to pin models?

No. Signed-out visitors can pin; the pins are kept for that browser and move to your account when you sign in.

Where did Claude Opus 5 and GPT-5.5 go?

Under Show older models, at the end of the menu. They still run and keep their leaderboard history. Typing their name in the open menu jumps to them too.

How do I compare two models on the same store?

Run a task once, then use Same task, another model on the run's page. Runs opens with the same store and task and the model you pick, ready to send.

Why can't I pick DeepSeek R1 or o4-mini any more?

They were retired on 14 November 2026 because the upstream model provider no longer serves them. Their runs stay on the leaderboard and their model pages.

What does "new" mean in the model menu?

The model was added to Playground in the last 30 days. It has fewer runs behind it, so its leaderboard figures are still filling in.



About UCP Checker

UCP Checker is the independent validation and observability layer for the Universal Commerce Protocol. We crawl, validate and grade every public UCP manifest we can find, run the merchant directory, the UCP Score, live adoption stats, the vertical guides and the authority-bound vendor map, and track the spec as it evolves so you don't have to — measured from two vantages (what a business declares, and what actually happens when an agent transacts), the same way for everyone, without picking winners.

Sources

Related coverage: Which AI models can actually complete a checkout? The benchmark refreshed · UCP Playground 0.24: Runs, Inspect and History in one app · UCP Playground now runs personal agents

What does a UCP agent find at your store?
Check it once and it gets its own status page.
Check a store

Keep reading

4 Nov 2026 UCP Playground Now Runs Personal Agents with Anthropic's Shopping Skills Product · 12 min 26 Oct 2026 UCP Location Validator: test whether an agent can find your shops Product · 7 min 22 Oct 2026 WebMCP runs in UCP Playground: watch an agent shop a store's website Product · 8 min