Models Reference

kicode supports any model available through OpenRouter. This page covers recommended models and how to choose the right one.

Default Model

The default model is:

x-ai/grok-code-fast-1

This was chosen for its balance of speed, cost, and coding capability.

Recommended Models

For Coding Tasks

Model	ID	Strengths
Claude 3.5 Sonnet	`anthropic/claude-3.5-sonnet`	Best overall coding, great reasoning
GPT-4 Turbo	`openai/gpt-4-turbo`	Broad knowledge, long context
Grok Code Fast	`x-ai/grok-code-fast-1`	Fast, cost-effective
DeepSeek Coder	`deepseek/deepseek-coder`	Specialized for code

For Complex Reasoning

Model	ID	Strengths
Claude 3 Opus	`anthropic/claude-3-opus`	Most capable, best reasoning
GPT-4	`openai/gpt-4`	Strong general reasoning

For Speed/Cost

Model	ID	Strengths
Claude 3 Haiku	`anthropic/claude-3-haiku`	Very fast, low cost
GPT-3.5 Turbo	`openai/gpt-3.5-turbo`	Fast, cheap

Selecting a Model

Via CLI Flag

kicode --model anthropic/claude-3.5-sonnet

Via Environment Variable

export KICODE_MODEL="anthropic/claude-3.5-sonnet"
kicode

Via Config File

model = "anthropic/claude-3.5-sonnet"

Model ID Format

OpenRouter model IDs follow the format:

provider/model-name

Examples:

anthropic/claude-3.5-sonnet
openai/gpt-4-turbo
google/gemini-pro
meta-llama/llama-3-70b-instruct

Finding Available Models

Browse all available models at openrouter.ai/models.

Filter by:

Modality: Text (required for kicode)
Context Length: Longer is better for large codebases
Pricing: Cost per token

Model Comparison

Capability Tiers

Tier	Examples	Best For
Flagship	Claude 3 Opus, GPT-4	Complex architecture, subtle bugs
Balanced	Claude 3.5 Sonnet, GPT-4 Turbo	Day-to-day coding
Fast	Claude 3 Haiku, GPT-3.5	Quick tasks, simple edits

Context Windows

Model	Context Length
GPT-4 Turbo	128k tokens
Claude 3.5 Sonnet	200k tokens
Claude 3 Opus	200k tokens

Longer context = can handle more code at once.

Pricing Considerations

Costs vary significantly. Check current pricing at openrouter.ai/models.

General guidance:

Opus/GPT-4: Higher cost, use for complex tasks
Sonnet/Turbo: Moderate cost, good default
Haiku/3.5: Low cost, good for simple tasks

Choosing the Right Model

Use Claude 3.5 Sonnet When:

Writing new features
Debugging complex issues
Refactoring code
Understanding unfamiliar codebases

Use Claude 3 Opus When:

Designing system architecture
Solving subtle logic bugs
Need highest accuracy

Use Grok Code Fast When:

Quick edits and fixes
Simple code generation
Cost-conscious usage

Use Claude 3 Haiku When:

Very simple tasks
High-volume operations
Maximum speed needed

Switching Models Mid-Session

You cannot switch models during a session. To use a different model:

Exit the current session
Start a new session with the desired model

# Exit current session
you: exit

# Start with new model
kicode --model anthropic/claude-3-opus

Troubleshooting

”Unknown model” Error

The model ID is not recognized by OpenRouter.

Fix: Verify the model ID at openrouter.ai/models.

”Rate limited” Error

You’ve exceeded the model’s rate limit.

Fix: Wait a few seconds or try a different model.

”Insufficient credits” Error

Your OpenRouter account needs more credits.

Fix: Add credits at openrouter.ai/credits.

Configuration Reference - Setting default model
Environment Variables - KICODE_MODEL variable