Skip to main content
Gemini CLI gives you access to Google’s most advanced language models. Understanding the different models and when to use them will help you get the best results for your tasks.

​
Model Selection

Use the /model command to configure which model Gemini CLI uses: 
/model
This opens a dialog with your model options. You can also use the --model flag when starting Gemini CLI: 
gemini --model gemini-2.5-flash
The /model command and --model flag do not override the model used by sub-agents. You may see other models in your usage reports even when specifying a particular model.

​
Model Options

Gemini CLI offers three main approaches to model selection:
Let the system automatically choose the best Gemini 3 model for your task.Available models:
  • gemini-3-pro-preview - For complex reasoning tasks
  • gemini-3-flash-preview - For fast, simple operations
Best for:
  • Most users and general-purpose development
  • Projects with a mix of complex and simple tasks
  • When you want optimal balance of speed and intelligence
Example use cases:
  • Building a web application (architecture planning + CSS generation)
  • Debugging issues (complex analysis + quick file edits)
  • Code reviews (deep understanding + formatting fixes)

​
Model Families

​
Pro Models

Pro models offer the highest levels of reasoning and creativity.

When to Use Pro

  • Complex multi-stage debugging
  • Architectural design and planning
  • Advanced code refactoring
  • Deep codebase analysis
  • Novel problem-solving requiring creativity
Characteristics:
  • Higher reasoning capabilities
  • Better for complex tasks
  • Slower response times
  • Higher token costs

​
Flash Models

Flash models provide fast responses for simpler tasks.

When to Use Flash

  • Simple code generation
  • Quick file edits
  • Format conversions (JSON to YAML)
  • Basic questions and explanations
  • Rapid iteration on small changes
Characteristics:
  • Very fast response times
  • Optimized for simple tasks
  • Lower token costs
  • Good for high-volume operations
Auto mode intelligently selects between Pro and Flash based on task complexity.

Why Use Auto

  • System automatically matches model to task complexity
  • Optimal balance of speed and intelligence
  • Cost-effective for mixed workloads
  • No manual switching required
How it works:
  • Simple tasks automatically use Flash for speed
  • Complex tasks automatically use Pro for quality
  • The system learns from task patterns
  • You get the best model for each specific request

​
Model Context Windows

Gemini 3 models feature a 1M token context window, allowing you to:
  • Work with extremely large codebases
  • Maintain very long conversations
  • Include extensive documentation in context
  • Process large files without truncation
Use /stats model to check your current token usage and see how much of the context window you’re using.

​
Best Practices

​
Default to Auto

For most users, the Auto option provides the best experience: 
/model
# Select: Auto (Gemini 3)
Benefits:
  • Automatically optimizes for each task
  • Balances speed and quality
  • Handles mixed workloads efficiently
  • Reduces cognitive overhead of model selection

​
Switch to Pro for Better Results

If Auto mode isn’t giving you the results you need: 
/model
# Select: Manual > gemini-3-pro-preview
Use Pro when:
  • Debugging complex, multi-component issues
  • Designing system architecture
  • Reverse engineering unfamiliar code
  • Solving novel problems without clear patterns
  • Working with complex business logic
Pro models are slower and use more quota. Only switch to Pro when you truly need the additional reasoning power.

​
Switch to Flash for Speed

For simple, repetitive tasks that need quick responses: 
/model
# Select: Manual > gemini-3-flash-preview
Use Flash when:
  • Converting between data formats
  • Generating boilerplate code
  • Making simple text edits
  • Answering straightforward questions
  • Performing bulk, simple operations

​
Model Configuration

You can configure your default model in several ways:

​
Command-Line Flag

Specify a model when launching: 
gemini --model gemini-2.5-flash

​
Environment Variable

Set a default in your shell profile: 
export GEMINI_MODEL="gemini-2.5-pro"

​
Settings File

Configure in ~/.gemini/settings.json: 
{
  "model": {
    "name": "auto"
  }
}

​
Precedence Order

When multiple configurations exist, they’re applied in this order:
  1. --model flag (highest priority)
  2. GEMINI_MODEL environment variable
  3. model.name in settings.json
  4. Default (auto)

​
Model Fallback

Gemini CLI includes automatic model fallback for resilience:
1

Model Failure Detected

If your selected model fails (quota exceeded, rate limiting, server errors), the CLI detects this automatically.
2

User Confirmation

You’re prompted to switch to a fallback model (unless configured for silent fallback).
3

Automatic Switch

If approved, the CLI uses an available fallback model to continue your session without interruption.
Internal utility calls (prompt completion, classification) use silent fallback: gemini-2.5-flash-lite → gemini-2.5-flash → gemini-2.5-pro without changing your configured model.

​
Model Capabilities

All Gemini models support:

Multimodal Input

Process text, images, PDFs, and audio files as input

Tool Calling

Execute tools for file operations, shell commands, and web access

Long Context

Handle up to 1M tokens in Gemini 3 models

Code Generation

Generate, analyze, and modify code across multiple languages

​
Quota and Pricing

​
Free Tier (Google Login)

When using Google Login (OAuth):
  • 60 requests/minute
  • 1,000 requests/day
  • Access to Gemini 3 models with 1M token context
  • No API key management required

​
Gemini API Key

  • 1,000 requests/day (free tier)
  • Mix of Flash and Pro models
  • Usage-based billing available for higher limits
  • Model-specific pricing applies

​
Vertex AI

  • Enterprise features and compliance
  • Scalable with billing account
  • Higher rate limits
  • Integration with Google Cloud
Check your current usage with /stats model to see requests, tokens, and quota information.

​
Next Steps

How It Works

Understand the architecture and request flow

Tools

Learn about available tools and how to use them

Configuration

Explore all configuration options

Authentication

Set up different authentication methods

Build docs developers (and LLMs) love

Get started for freeTalk to us