Skip to content
clauderon

Model Selection

Clauderon supports 22 different AI models across three major providers: Claude (Anthropic), GPT/Codex (OpenAI), and Gemini (Google). This guide helps you choose the right model for your use case.

Overview

Different models offer different trade-offs:

  • Capability - Complex reasoning, code generation, refactoring
  • Speed - Response latency and throughput
  • Cost - Token pricing and context window efficiency
  • Context - Maximum input/output size
  • Features - Plan mode, tool use, multi-modal support

Model Families

Claude Models (Anthropic)

7 Claude models available, powered by Anthropic’s latest technology:

ModelCapabilitySpeedContextUse Cases
Claude Opus 4.5HighestModerate200KComplex refactoring, architecture design
Claude Sonnet 4.5HighFast200KDefault - balanced for most tasks
Claude Haiku 4.5ModerateFastest200KQuick edits, simple tasks, iterations
Claude Opus 4.1High (Agentic)Moderate200KMulti-step reasoning, planning
Claude Opus 4HighModerate200KPrevious generation, still capable
Claude Sonnet 4ModerateFast200KPrevious generation, fast
Claude Haiku 4LowerFastest200KPrevious generation, simple tasks

Model IDs:

  • claude-opus-4-5 or claude-opus-4-5-20251101
  • claude-sonnet-4-5 or claude-sonnet-4-5-20250929 (default)
  • claude-haiku-4-5 or claude-haiku-4-5-20250701
  • claude-opus-4-1 or claude-opus-4-1-20241129
  • claude-opus-4 or claude-opus-4-20240229
  • claude-sonnet-4 or claude-sonnet-4-20240229
  • claude-haiku-4 or claude-haiku-4-20240307

Features:

  • ✅ Plan mode supported (all models)
  • ✅ Tool use (function calling)
  • ✅ Multi-modal (image understanding)
  • ✅ Extended thinking (Opus models)
  • ✅ Code generation optimized

Authentication:

  • Requires Anthropic API key or OAuth

GPT/Codex Models (OpenAI)

10 GPT models available, including specialized Codex variants:

ModelCapabilitySpeedContextUse Cases
GPT-5.2-CodexHighest (Code)Moderate128KDefault GPT - optimized for code
GPT-5.2HighestModerate128KGeneral tasks, reasoning
GPT-5.2-InstantHighFastest128KQuick responses, iterations
GPT-5.2-ThinkingHighestSlowest128KComplex reasoning, deep analysis
GPT-5.2-ProHighest+Slow128KMost capable, expensive
GPT-5.1HighFast128KPrevious generation, balanced
GPT-5.1-InstantModerateFastest128KPrevious generation, fast
GPT-5.1-ThinkingHighSlow128KPrevious generation, reasoning
GPT-4.1ModerateFast8KOlder generation, lower cost
o3-miniHigh (Reasoning)Moderate128KSpecialized reasoning model

Model IDs:

  • gpt-5.2-codex (default for GPT)
  • gpt-5.2
  • gpt-5.2-instant
  • gpt-5.2-thinking
  • gpt-5.2-pro
  • gpt-5.1
  • gpt-5.1-instant
  • gpt-5.1-thinking
  • gpt-4.1
  • o3-mini

Features:

  • ✅ Tool use (function calling)
  • ✅ Code generation (especially Codex variants)
  • ⚠️ Plan mode (limited support)
  • ✅ Multi-modal (GPT-5.2 models)
  • ✅ Thinking mode (Thinking variants)

Authentication:

  • Requires OpenAI API key

Gemini Models (Google)

5 Gemini models available, featuring very large context windows:

ModelCapabilitySpeedContextUse Cases
Gemini 3 ProHighestModerate1MHuge codebases, large context
Gemini 3 FlashHighFastest1MFast iterations with large context
Gemini 2.5 ProHighModerate1MPrevious generation, still capable
Gemini 2.0 FlashModerateFastest1MPrevious generation, fast
Gemini 2.0 Flash-ThinkingHighSlow1MReasoning with large context

Model IDs:

  • gemini-3-pro
  • gemini-3-flash (default for Gemini)
  • gemini-2.5-pro
  • gemini-2.0-flash
  • gemini-2.0-flash-thinking

Features:

  • ✅ Very large context (1M tokens)
  • ✅ Tool use
  • ✅ Multi-modal
  • ⚠️ Plan mode (limited support)
  • ✅ Flash models for speed

Authentication:

  • Requires Google AI API key

Selecting Models

Via Web UI

The Web UI provides full model selection:

  1. Create New Session

    • Click “New Session”
    • Configure repository and backend

  2. Advanced Options

    • Expand “Advanced Options” section
    • Find “Model Override” dropdown

  3. Choose Model

    • Browse all 22 available models
    • Models grouped by provider (Claude, GPT, Gemini)
    • Default model shown (Claude Sonnet 4.5)

  4. Create Session

    • Model applies to this session only
    • Can be changed by recreating session

Via API

Specify model in session creation request:

Terminal window
curl -X POST http://localhost:3030/api/sessions \
  -H "Content-Type: application/json" \
  -d '{
    "name": "my-session",
    "repository": "/path/to/repo",
    "backend": "docker",
    "agent": "claude-code",
    "model": "claude-opus-4-5"
  }'

Model parameter:

  • Optional (defaults to claude-sonnet-4-5)
  • Must be valid model ID
  • Case-sensitive

Via CLI

Status: Not yet implemented

The CLI currently does not support model selection. Model defaults to Claude Sonnet 4.5.

Workaround:

  • Create session via Web UI or API with desired model
  • Attach to session from CLI:
    Terminal window
    clauderon attach my-session

Via TUI

Status: Not yet implemented

The TUI currently does not support model selection during session creation.

Workaround:

  • Create session via Web UI or API with desired model
  • Attach to session from TUI:
    Terminal window
    clauderon tui
    # Select and attach to pre-created session

Model Capabilities Comparison

Plan Mode Support

Plan mode allows the agent to create an implementation plan before executing:

ProviderPlan Mode Support
Claude✅ Full support (all models)
GPT⚠️ Limited support (best with Thinking models)
Gemini⚠️ Limited support

Best for plan mode: Claude Opus 4.5, Claude Sonnet 4.5

Context Window Sizes

ProviderContext Window
Claude200K tokens
GPT128K tokens (most), 8K (GPT-4.1)
Gemini1M tokens

Best for large codebases: Gemini 3 Pro, Gemini 3 Flash

Code Generation

All models support code generation, but some are optimized:

  • Best: GPT-5.2-Codex (specialized for code)
  • Excellent: Claude Opus 4.5, Claude Sonnet 4.5
  • Very Good: GPT-5.2, Gemini 3 Pro
  • Good: All other models

Speed

Fastest to slowest (approximate):

  1. Fastest: Claude Haiku 4.5, GPT-5.2-Instant, Gemini 3 Flash
  2. Fast: Claude Sonnet 4.5, GPT-5.1, Gemini 2.0 Flash
  3. Moderate: Claude Opus 4.5, GPT-5.2, Gemini 3 Pro
  4. Slow: GPT-5.2-Thinking, Gemini 2.0 Flash-Thinking
  5. Slowest: GPT-5.2-Pro

Cost

Approximate relative costs (actual pricing varies by provider):

  • Most Expensive: GPT-5.2-Pro, Claude Opus 4.5
  • Expensive: GPT-5.2, Claude Opus 4.1, Gemini 3 Pro
  • Moderate: Claude Sonnet 4.5, GPT-5.1, Gemini 2.5 Pro
  • Affordable: GPT-5.1-Instant, Gemini 3 Flash
  • Economical: Claude Haiku 4.5, GPT-4.1, Gemini 2.0 Flash

Check provider pricing pages for exact rates.

Backend Compatibility

All models work with all backends:

  • ✅ Docker - Full support
  • ✅ Kubernetes - Full support
  • ✅ Zellij - Full support
  • ✅ Sprites - Full support
  • ✅ Apple Container - Full support

Model selection is independent of backend choice.

Authentication Requirements

API Keys

Most models require API keys:

Claude models:

Terminal window
export ANTHROPIC_API_KEY="sk-ant-..."

GPT models:

Terminal window
export OPENAI_API_KEY="sk-..."

Gemini models:

Terminal window
export GOOGLE_API_KEY="AIza..."

OAuth (Claude Only)

Claude models can also use OAuth:

Terminal window
clauderon auth login
# Opens browser for OAuth flow

OAuth tokens are stored securely and automatically refreshed.

Choosing the Right Model

For Most Tasks (Default)

Claude Sonnet 4.5 - Best balance of capability, speed, and cost

{
  "model": "claude-sonnet-4-5"
}

For Complex Refactoring

Claude Opus 4.5 - Highest capability, best for architecture and complex changes

{
  "model": "claude-opus-4-5"
}

For Quick Iterations

Claude Haiku 4.5 or GPT-5.2-Instant - Fastest responses

{
  "model": "claude-haiku-4-5"
}

For Large Codebases

Gemini 3 Pro - 1M token context window

{
  "model": "gemini-3-pro"
}

For Code Generation

GPT-5.2-Codex - Optimized for coding tasks

{
  "model": "gpt-5.2-codex"
}

For Deep Reasoning

GPT-5.2-Thinking or Claude Opus 4.1 - Extended thinking capabilities

{
  "model": "gpt-5.2-thinking"
}

For Cost Optimization

Claude Haiku 4.5 or Gemini 2.0 Flash - Most economical

{
  "model": "claude-haiku-4-5"
}

Model Selection Strategies

Development Workflow

Use different models at different stages:

  1. Exploration - Claude Sonnet 4.5 (balanced)
  2. Implementation - GPT-5.2-Codex (code optimized)
  3. Refactoring - Claude Opus 4.5 (complex changes)
  4. Testing/Debugging - Claude Haiku 4.5 (quick iterations)

Task-Based Selection

Bug fixes: Claude Haiku 4.5 (fast, good enough) New features: Claude Sonnet 4.5 (balanced) Architecture design: Claude Opus 4.5 (complex reasoning) Code review: GPT-5.2-Codex (code understanding) Documentation: Claude Sonnet 4.5 (good writing) Large codebase navigation: Gemini 3 Pro (huge context)

Budget Optimization

High budget: Use Claude Opus 4.5 or GPT-5.2-Pro for everything Medium budget: Use Claude Sonnet 4.5 as default, Opus for complex tasks Low budget: Use Claude Haiku 4.5 or Gemini Flash, Sonnet only when needed

Speed Optimization

Need fastest responses: Claude Haiku 4.5, GPT-5.2-Instant, Gemini 3 Flash Balanced speed/quality: Claude Sonnet 4.5, GPT-5.2 Quality over speed: Claude Opus 4.5, GPT-5.2-Thinking

Model Switching

During Session

Models cannot be changed mid-session. To switch models:

  1. Recreate session with new model:

    Terminal window
    # Via Web UI: Session → Recreate → Choose new model
    # Or create new session

  2. Session state preserved (if using “Recreate” option)

    • Chat history maintained
    • Git state preserved
    • Metadata retained

Multi-Session Strategy

Run multiple sessions with different models:

Terminal window
# Session 1: Exploration (Sonnet)
clauderon create --model claude-sonnet-4-5 exploration


# Session 2: Complex refactor (Opus)
clauderon create --model claude-opus-4-5 refactor


# Session 3: Quick fixes (Haiku)
clauderon create --model claude-haiku-4-5 fixes

Best Practices

  1. Start with default - Claude Sonnet 4.5 works for most tasks
  2. Upgrade for complexity - Use Opus when Sonnet struggles
  3. Downgrade for speed - Use Haiku for simple, quick tasks
  4. Match context to codebase - Use Gemini for very large projects
  5. Consider cost - Monitor token usage and adjust accordingly
  6. Test different models - Different models excel at different tasks
  7. Use plan mode - Leverage Claude’s plan mode for complex implementations
  8. Codex for pure coding - GPT-5.2-Codex for heavy code generation
  9. Thinking for reasoning - Use Thinking variants for complex logic
  10. Flash for iteration - Gemini Flash or GPT Instant for rapid feedback

Troubleshooting

Model Not Available

Error: Model 'xyz' not found

Causes:

  • Typo in model ID
  • Model not supported by Clauderon
  • Model deprecated or renamed

Solution:

  • Check model ID spelling (case-sensitive)
  • Refer to model list above
  • Use default model if unsure

Authentication Failed

Error: Authentication failed for model

Causes:

  • API key not set
  • API key invalid or expired
  • Wrong provider for model (e.g., OpenAI key for Claude model)

Solution:

Terminal window
# Check environment variables
echo $ANTHROPIC_API_KEY
echo $OPENAI_API_KEY
echo $GOOGLE_API_KEY


# Set missing key
export ANTHROPIC_API_KEY="sk-ant-..."


# Or use OAuth (Claude only)
clauderon auth login

Rate Limiting

Error: Rate limit exceeded

Causes:

  • Too many requests to provider API
  • Account limits reached
  • Concurrent session limit

Solutions:

  • Wait and retry
  • Upgrade API tier with provider
  • Use different model/provider
  • Reduce concurrent sessions

Poor Quality Responses

Problem: Model produces low-quality code or explanations

Solutions:

  • Try a more capable model (e.g., Opus instead of Haiku)
  • Use Codex variant for code tasks
  • Enable plan mode for complex tasks
  • Provide more context in prompt
  • Break task into smaller steps

Slow Responses

Problem: Model takes too long to respond

Solutions:

  • Use faster model (Haiku, Instant, Flash)
  • Reduce context size (smaller prompts)
  • Use streaming (enabled by default)
  • Check network latency to provider

Context Window Exceeded

Error: Context window exceeded

Causes:

  • Codebase too large for model’s context
  • Too much chat history
  • Large file attachments

Solutions:

  • Use Gemini models (1M context)
  • Archive old sessions (clear history)
  • Work with smaller portions of codebase
  • Remove large files from context

Model-Specific Tips

Claude Opus 4.5

  • Best for plan mode (creates detailed implementation plans)
  • Excellent at refactoring entire codebases
  • Use for architecture decisions
  • Higher cost, reserve for complex tasks

Claude Sonnet 4.5

  • Default for good reason - balanced all-around
  • Fast enough for iterative development
  • Capable enough for most tasks
  • Best cost/performance ratio

Claude Haiku 4.5

  • Use for quick edits and simple tasks
  • Great for testing and debugging loops
  • Much faster than Opus/Sonnet
  • Lowest cost option for Claude

GPT-5.2-Codex

  • Specialized for code generation
  • Better at boilerplate and repetitive code
  • May be less creative than Claude for architecture
  • Good for pure coding tasks

Gemini 3 Pro

  • Use when you need 1M context window
  • Great for exploring large codebases
  • Can handle entire monorepos in context
  • Slower than Flash variants

GPT-5.2-Thinking

  • Use for complex algorithmic problems
  • Shows reasoning steps (helpful for learning)
  • Slower but more thorough
  • Good for debugging complex issues

See Also