Google / Model reference

Gemini 3 Pro

Name: Gemini 3 Pro
Author: Google

The best model in the world for multimodal understanding, and Google's most powerful agentic and vibe-coding model yet, delivering richer visuals and deeper interactivity, all built on a foundation of state-of-the-art reasoning.

See the code

Model IDgemini-3-pro-preview

What it can do.

Streaming

celeste.text.generateGenerate text

celeste.images.analyzeUnderstand media

celeste.audio.analyzeUnderstand media

celeste.videos.analyzeUnderstand media

Parameters.

Supported controls from Celeste’s model registry. Values and availability differ between models.

Gemini 3 Pro parameters and constraints
Parameter	Accepted values
`temperature`	0 – 2
`max_tokens`	1 – 65536
`thinking_level`	`lowhigh`
`tools`	List of tool definitions
`output_schema`	Pydantic BaseModel
`image`	Image artifacts
`video`	videos
`audio`	audio

Start with a few lines

Put it to work.

Use your provider credentials. Call the model through the SDK and build around its output.

import celeste
response = await celeste.text.generate(
    model="gemini-3-pro-preview",
    prompt="Write a project brief for a neighborhood repair café."
)
print(response.content)

This example uses text.generate. Explore the quickstart for installation and authentication.