Gemini Evals - Phoenix

Google Gemini via Vertex AI#

Use the LLM wrapper with the "vertex" provider to run evals with Gemini models. Authenticate by setting your project and location, or by passing credentials directly.

pip install "arize-phoenix-evals>=3" google-cloud-aiplatform

import osfrom phoenix.evals import LLMos.environ["CLOUD_ML_PROJECT_ID"] = "my-project-id"os.environ["CLOUD_ML_REGION"] = "us-central1"llm = LLM(provider="vertex", model="gemini-2.5-flash")llm.generate_text(prompt="Hello there, how are you?")# Output: "Hello! I'm doing well, thank you for asking!"

Google Gemini via Google AI Studio#

Alternatively, use the "google" provider with a Google AI Studio API key:

pip install "arize-phoenix-evals>=3" google-genai

import osfrom phoenix.evals import LLMos.environ["GOOGLE_API_KEY"] = "<your-api-key>"llm = LLM(provider="google", model="gemini-2.5-flash")llm.generate_text(prompt="Hello there, how are you?")# Output: "Hello! I'm doing well, thank you for asking!"