Model Selection

Git-Iris uses a dual-model strategy: primary models for complex analysis and fast models for simple tasks.

Model Strategy

Model Type	Used For	Examples
Primary	Commit messages, code reviews, PR descriptions	claude-sonnet-4-5-20250929, gpt-5.1
Fast	Status updates, parsing, simple queries	claude-haiku-4-5-20251001, gpt-5.1-mini

This dual-model approach optimizes for both quality and speed.

Default Models by Provider

OpenAI

[providers.openai]
model = "gpt-5.1"
fast_model = "gpt-5.1-mini"

Model	Use Case	Context	Notes
gpt-5.1	Primary	128K	Best for complex analysis
gpt-5.1-mini	Fast	128K	Quick status updates

Anthropic

[providers.anthropic]
model = "claude-sonnet-4-5-20250929"
fast_model = "claude-haiku-4-5-20251001"

Model	Use Case	Context	Notes
claude-sonnet-4-5-20250929	Primary	200K	Excellent for code analysis
claude-haiku-4-5-20251001	Fast	200K	Fastest response times

Google

[providers.google]
model = "gemini-3-pro-preview"
fast_model = "gemini-2.5-flash"

Model	Use Case	Context	Notes
gemini-3-pro-preview	Primary	1M	Largest context window
gemini-2.5-flash	Fast	1M	Good for large diffs

Configuring Models

Via CLI

# Set primary model
git-iris config --provider anthropic --model claude-sonnet-4-5-20250929

# Set fast model
git-iris config --provider anthropic --fast-model claude-haiku-4-5-20251001

Via Config File

[providers.anthropic]
api_key = "sk-ant-..."
model = "claude-sonnet-4-5-20250929"
fast_model = "claude-haiku-4-5-20251001"

When to Use Which Model

Primary Model Tasks

• Commit message generation — Needs context understanding
• Code reviews — Requires deep analysis
• PR descriptions — Synthesizes multiple changes
• Changelogs — Categorizes and summarizes
• Release notes — Produces polished documentation

Fast Model Tasks

• Status updates — "Analyzing file 3 of 15..."
• Progress parsing — Extracting structured data
• Tool responses — Simple confirmations
• Chat queries — Quick interactions

Model Selection Criteria

Choose Primary Model Based On

Priority	Consideration	Recommendation
Quality	Need best analysis	Claude Sonnet 4.5
Context	Large changesets (>50 files)	Gemini 3 Pro (1M tokens)
Speed	Fast turnaround	GPT-5.1
Cost	Budget constraints	Use fast models more

Optimize Fast Model For

• Response time — Haiku, GPT-5.1-mini, Gemini Flash
• Context window — All fast models support large context
• Availability — Check provider rate limits

Custom Model Configuration

You can use any model supported by your provider:

# OpenAI custom model
git-iris config --provider openai --model gpt-4o

# Anthropic custom model
git-iris config --provider anthropic --model claude-opus-4-5

# Google custom model
git-iris config --provider google --model gemini-2.0-flash-exp

Model Fallback Behavior

If a configured model is unavailable, Git-Iris will:

1. Attempt to use the provider's default model
2. Report an error with the model name
3. Suggest checking provider documentation

Context Window Management

Git-Iris automatically manages context to fit within model limits:

Scenario	Strategy
Small changeset (<10 files)	Full context
Medium changeset (10-20 files)	Relevance scoring
Large changeset (20+ files)	Parallel subagent analysis

Override token limits per provider:

git-iris config --provider anthropic --token-limit 150000

Model Performance Tips

For Large Repositories

[providers.google]
model = "gemini-3-pro-preview"  # 1M context window

For Speed-Critical Workflows

[providers.anthropic]
model = "claude-haiku-4-5-20251001"  # Fast even for primary tasks
fast_model = "claude-haiku-4-5-20251001"

For Maximum Quality

[providers.anthropic]
model = "claude-sonnet-4-5-20250929"  # Best code understanding

Monitoring Model Usage

Enable debug mode to see which model handles each task:

git-iris gen --debug

Output shows:

• Model name
• Token usage
• Tool calls
• Response time

Cost Optimization

Minimize Costs

# Use fast model for everything
[providers.openai]
model = "gpt-5.1-mini"
fast_model = "gpt-5.1-mini"
token_limit = 8000  # Lower limit

Balance Quality and Cost

# Standard setup
[providers.anthropic]
model = "claude-sonnet-4-5-20250929"  # Quality for commits
fast_model = "claude-haiku-4-5-20251001"  # Speed for status

Troubleshooting

Issue	Solution
"Model not found"	Check provider documentation for available models
Slow responses	Switch to a faster model
Context exceeded	Reduce token_limit or use model with larger window
Poor quality	Use a more capable primary model

Model Comparison Table

Model	Provider	Context	Speed	Quality	Cost
claude-sonnet-4-5	Anthropic	200K	Medium	Excellent	Medium
claude-haiku-4-5	Anthropic	200K	Fast	Good	Low
gpt-5.1	OpenAI	128K	Medium	Excellent	Medium
gpt-5.1-mini	OpenAI	128K	Fast	Good	Low
gemini-3-pro	Google	1M	Slow	Excellent	High
gemini-2.5-flash	Google	1M	Fast	Good	Low