Choosing a Model (Gemini)

Testronaut supports Google Gemini 2.5 models for agentic (tool-using) workflows.
You can select a Gemini model during testronaut --init, store it in testronaut-config.json, or override it per run via CLI or environment variables.

Heads-up: Gemini model access and rate limits depend on your Google Cloud or MakerSuite account.
If you select a model your account doesn’t have access to, the API will respond with a 404 or PERMISSION_DENIED error.

---

Supported Gemini model families

All of the following are compatible with Testronaut’s tool calling & DOM agent features:

• Gemini 2.5 Pro — High-quality, balanced model for reasoning and code
• Gemini 2.5 Flash — Fast, low-latency model optimized for throughput
• Gemini 2.5 Flash-8B — Lightweight Flash variant for cost efficiency

These identifiers map directly to Google’s Generative Language API model IDs (e.g., gemini-2.5-pro).
Testronaut forwards your chosen ID directly to the API.

---

How Testronaut decides which model to use

1. CLI override (highest precedence)

   testronaut --model gemini-2.5-pro

2. Environment override

   export TESTRONAUT_MODEL=gemini-2.5-flash

3. Project config
   Stored in your testronaut-config.json after initialization.

4. Fallback
   Defaults to gemini-2.5-flash if none are specified.

The active provider/model appear in:

• CLI JSON and HTML reports (results.llm.provider & results.llm.model)
• The Mission Control companion app (header and table pills)

---

Example configuration

testronaut-config.json

{
  "initialized": true,
  "provider": "gemini",
  "model": "gemini-2.5-flash",
  "outputDir": "missions/mission_reports",
  "projectName": "your-project",
  "maxTurns": 20
}

Override per run:

testronaut checkoutFlow.mission.js --model gemini-2.5-pro

Or for CI/CD or temporary overrides:

TESTRONAUT_MODEL=gemini-2.5-flash-8b testronaut

---

Cost / performance guide (quick picks)

Model	Strengths	Typical Use	Trade-offs
Gemini 2.5 Pro	Strong reasoning and comprehension	Complex workflows, DOM reasoning, deeper analysis	Slower, higher cost
Gemini 2.5 Flash	Very fast, low latency	Iterative development, large suites, smoke tests	Slightly shallower reasoning
Gemini 2.5 Flash-8B	Lightweight, cost-effective	Budget-conscious or parallel testing	Less capable in nuanced reasoning

---

Recommendations

Fast & low cost: gemini-2.5-flash
Best general agent: gemini-2.5-pro
High concurrency (CI): gemini-2.5-flash-8b

---

Rate limits & backoff

Testronaut applies token-aware throttling and will automatically back off if you approach your per-minute token limits.
Unlike OpenAI, Gemini’s rate limits may not always be returned in headers — Testronaut uses conservative defaults and runtime heuristics to avoid throttling errors.

---

Troubleshooting

“404: model not found”
You may not have access to that Gemini model.
Check the available models under your API key or try switching to gemini-2.5-flash.

“429: rate limit exceeded”
Try using gemini-2.5-flash or gemini-2.5-flash-8b for high-volume or concurrent runs.

Model not listed?
Testronaut forwards model IDs directly, so newer releases (e.g. gemini-3.0) will work as long as they follow the same API schema.

---

Reports

Testronaut’s JSON and HTML reports include:

• llm.provider: "gemini"
• llm.model: e.g. "gemini-2.5-flash"

The Mission Control dashboard displays these as colored pills in the report tables.