Models

Access via Settings → Models in your project.

The Models settings page lets you define how BrowserStack AI Evals handles models it encounters in traces — including token counting, cost calculation, and display metadata.

Why Configure Models?

When the platform sees a model name in a trace (e.g., gpt-4o, claude-3-5-sonnet-20241022), it looks up:

• The tokenizer to use for counting tokens when usage metadata is missing
• The pricing to calculate estimated cost per trace
• A display name for the dashboard

Built-in model definitions are provided for common models. Add a custom definition if:

• You use a fine-tuned or self-hosted model not in the default list
• You want to override pricing for a model (e.g., you have a custom enterprise rate)
• You use an internal model identifier that doesn't match a known name

Adding a Custom Model

1. Go to Settings → Models
2. Click Add model
3. Fill in:
   • Model ID — the exact string your SDK sends in GenerationBody.model (must match)
   • Match pattern — optional regex to match a family of model versions (e.g., ^gpt-4o.*)
   • Display name — friendly name shown in the dashboard
   • Tokenizer — which tokenizer to use for token counting (see below)
   • Start date / End date — optional validity window for pricing changes
4. Set pricing (see below)
5. Click Save

Tokenizer Selection

The tokenizer is used to estimate token counts when usage data is not provided by the LLM API.

Pricing Configuration

Pricing is specified per 1,000 tokens (or per request for fixed-cost models).

Pricing appears in trace detail views and is aggregated in usage dashboards.

Service Tier Pricing

If your provider charges different rates for different service tiers (e.g., batch vs. real-time), you can create multiple model entries with the same model ID but different date ranges or match patterns to represent each tier.

Editing and Deleting Models

Click the model name in the list to edit its configuration. Use the delete icon to remove a custom model definition. Deleting a model definition does not delete traces — it only removes the cost and tokenizer lookup.

Permissions