Choosing Models

Choosing models

UCP Playground carries 35 models from 8 labs, each with different strengths. The right one depends on whether you care most about how far runs get, speed, or cost.

The model menu

Choose the model with the Model pill under the message box on Runs. It starts on Gemini 3.6 Flash, marked default.

  • Pinned -- your models come first, in the order you pinned them. Pin one with the pin on its row, or from Edit pinned models… at the end of the menu. Signed-out visitors can pin too; the pins come along when you sign in.
  • By lab -- every other model sits under its lab, A to Z, newest first. Models added in the last 30 days say new.
  • Older models -- earlier versions sit behind Show older models. They still run and keep their leaderboard history. Typing a model's name finds it either way.

On a run's page, Same task, another model opens Runs at the same store with the same task and the model you pick, ready to send -- the quickest way to compare models on one store.

Available models

  • Alibaba -- Qwen 3.8 Max, Qwen 3.8 Flash
  • Anthropic -- Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5
  • DeepSeek -- DeepSeek V4.1 Flash, DeepSeek V4 Pro
  • Google -- Gemini 4 Argon, Gemini 4 Hydrogen, Gemini 3.8 Flash, Gemini 3.6 Flash
  • Meta -- Muse Ember, Muse Spark 1.3
  • Moonshot AI -- Kimi K3.1 Pro
  • OpenAI -- GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Luna
  • xAI -- Grok 4.7

Older models (still selectable): Qwen 3.7 Max, Claude Opus 5, Claude Sonnet 5, Claude Opus 4.8, Claude Sonnet 4.6, DeepSeek V4 Flash, DeepSeek V3.2, Gemini 3.7 Flash, Gemini 3.5 Flash, Gemini 3.1 Pro, Gemini 2.5 Pro, Gemini 2.5 Flash, Llama 4 Maverick, Kimi K3, GPT-5.5, GPT-4o and Grok 4.5.

Retired models keep their leaderboard stats and detail pages but can no longer start new runs: DeepSeek R1 and o4-mini (no longer served upstream, 14 Nov 2026), QwQ 32B, Grok 3 Mini, Llama 3.3 70B and Grok 4.3. If a model you had chosen is retired, Runs starts on the default and says so once.

Understanding the trade-offs

Every model balances three factors: capability, speed, and cost (measured in token usage). More capable models tend to be slower and consume more tokens, but they handle complex shopping scenarios better -- especially variant selection, multi-step checkout flows, and recovery from unexpected tool responses.

Flagship models

Claude Opus 5.5 is Anthropic's latest and most capable model. The Opus line excels at reasoning through complex catalogs, correctly matching variants (size, color, style), and navigating multi-step checkouts. The trade-off is higher token usage and longer runs. Use it when you want the furthest-reaching runs regardless of cost.

GPT-6.1 Sol is OpenAI's strong all-rounder that balances capability with efficiency. It handles most shopping scenarios well and uses fewer tokens than Opus. A solid default for benchmarking.

Claude Sonnet 5.5, Gemini 4 Argon, Grok 4.7, Kimi K3.1 Pro, Qwen 3.8 Max, DeepSeek V4 Pro and Muse Spark 1.3 are flagship alternatives with strong tool-calling abilities. Models marked new are still accumulating Playground benchmark data -- check the leaderboard for current numbers.

Balanced models

Claude Sonnet 4.6 and Gemini 2.5 Pro sit in the sweet spot between capability and cost. They handle the full shopping funnel reliably for most stores while keeping token budgets reasonable. They are good starting points for first tests. Both now sit under Show older models and still run.

Fast and lightweight models

The Gemini Flash line (Gemini 3.8 Flash, the newest, then 3.7, 3.6, 3.5, and 2.5 Flash) gives the fastest and cheapest options. They finish runs quickly with few tokens, making them ideal for smoke tests, regression checks, and high-volume benchmarking. They may struggle with complex variant matching or stores that return unusual response formats.

GPT-6 Luna (OpenAI's low-cost GPT-6 tier), Qwen 3.8 Flash, DeepSeek V4.1 Flash, DeepSeek V4 Flash, DeepSeek V3.2, GPT-4o, and Llama 4 Maverick offer budget-friendly alternatives with varying strengths. They work well for straightforward catalogs and simple shopping queries.

Recommendations

If you are not sure where to start:

  • First-time testing -- start with the default, Gemini 3.6 Flash, then pin the models you want to compare
  • Quick smoke tests -- use a small model (Gemini 3.6 or 3.8 Flash, Gemini 4 Hydrogen, Claude Haiku 5.5, GPT-6 Luna) for fast, inexpensive runs
  • Comparing models -- run a task once, then use Same task, another model on the run page
  • Comparing stores -- pick one model and use it consistently across stores so results are comparable
Tip

Check the Leaderboard for how far each model gets and its average token use across many runs on real stores. The Models pages have the detail per model.

Something wrong on this page?