BrowserStack AI Evals
EvaluationPlayground

Model Comparison

Open multiple Playground windows to compare models, temperatures, and prompt variants side by side.

Model Comparison

The Playground supports up to 5 windows open simultaneously on desktop. Each window is independent — it has its own model, parameters, and messages — so you can run direct A/B comparisons or test multiple variants in one click.

Setting Up a Side-by-Side Comparison

Open the Playground

Navigate to your project and click Playground in the sidebar. You start with one window.

Duplicate a window

Click Duplicate (the copy icon) on the window header to create a second window that inherits the current model, messages, and parameters. This is the fastest way to start an A/B test.

Alternatively, use Add Window from the window controls area to open a blank second window.

Configure each window independently

In each window you can change:

  • Model — switch provider and model (e.g. GPT-4o in one window, Claude Sonnet in another).
  • Model Parameters — expand the parameters panel to set temperature, max tokens, top-p, and other provider-specific settings independently per window.
  • Messages — adjust the system prompt or user message to test prompt variants.

Run all windows simultaneously

Click Run All in the page header (or press Cmd+Enter / Ctrl+Enter when multiple windows are open). All windows execute in parallel and their responses stream in side by side.

Click Stop All at any time to cancel all in-flight requests.

Compare outputs

Review responses side by side. Each window's output panel shows the generated text. When a dataset is connected, the results table columns correspond to each window's model, making cost, latency, and score comparisons straightforward.

Add more windows

Repeat the Duplicate step to add additional windows. You can have up to 5 windows open at once.

Use Reset Playground (the reset icon in the page header) to clear all windows and start fresh.

Closing a Window

Click the on the window header to remove that window. At least one window always remains open.

Mobile Behavior

On mobile screens only one window is shown at a time. Multi-window comparison is a desktop feature.