Skip to main content

Choosing a Model (Google Gemini)

Compatibility

Verified against Testronaut CLI v1.10.0. The interactive catalog is defined by the installed CLI, so run testronaut --init to see the models available in your version.

Testronaut supports Google Gemini models for agentic browser workflows, including tool calls and screenshots.

Setup​

Create or manage a key in Google AI Studio. Google's Gemini API key guide covers current key types, restrictions, and environment setup.

Select Google Gemini during initialization:

testronaut --init

Add your API key to .env:

GEMINI_API_KEY=AIza...

A new project defaults to the latest stable model in the CLI catalog:

testronaut-config.json
{
  "provider": "gemini",
  "model": "gemini-3.8-flash"
}

Current model catalog​

ModelBest for
gemini-3.8-flashRecommended stable model for general agentic missions
gemini-3.7-flashCompatibility with the previous stable generation
gemini-3.6-flashToken-efficient general testing
gemini-3.5-flash-liteFast, cost-conscious, high-volume runs
gemini-3.1-pro-previewComplex reasoning where preview-model use is acceptable

The CLI also lists gemini-2.5-pro, gemini-2.5-flash, and gemini-2.5-flash-lite as legacy choices.

caution

Model access, lifecycle status, rate limits, and pricing depend on your Google account and project. A model shown by Testronaut may still be unavailable to a particular API key.

Per-run overrides​

Pair the provider and model when temporarily switching from another provider:

testronaut checkout.mission.js --provider=gemini --model=gemini-3.8-flash

Environment overrides work too:

TESTRONAUT_PROVIDER=gemini TESTRONAUT_MODEL=gemini-3.5-flash-lite testronaut

Resolution order is:

  1. TESTRONAUT_PROVIDER and TESTRONAUT_MODEL
  2. provider and model in testronaut-config.json
  3. CLI defaults

The selected provider and model appear in local and uploaded reports.

Troubleshooting​

Model not found or permission denied: Confirm that the exact model ID is available to the API key and Google project.

Rate limit exceeded: Reduce concurrency, wait for the quota window to reset, or choose a model intended for higher-volume workloads.