Get Started

Models

Thesis supports OpenAI models for AI-powered features.

Supported Models

Thesis integrates with OpenAI's API through the @openai/agents framework. The following models are available:

ModelTypeBest For
gpt-4oStandardGeneral use, recommended default
o1-previewReasoningComplex multi-step problems
o3-miniReasoningFast reasoning tasks

Model Selection

You can specify the model when using the CLI:

thesis exec "Analyze this data" --model gpt-4o

The default model is configured in your Thesis settings. To change it:

thesis auth login

Reasoning Models

Reasoning models (o1 and o3 series) use chain-of-thought processing for complex problem-solving. However, they have important limitations:

Important

Reasoning models operate with limited tool access for optimal performance. When using o1 or o3 models, the agent cannot use tools like file editing or code execution.

For tasks that require tool use (modifying notebooks, executing code, searching the web), use gpt-4o instead.

Configuration

Model behavior is automatically adjusted based on:

  • Agent mode - Higher reasoning effort in Agent mode, lower in Ask mode
  • Message complexity - Longer or more complex prompts trigger higher reasoning effort
  • Task keywords - ML/data science keywords increase reasoning allocation

The reasoning effort levels are:

  • High - 2000-4000 tokens for complex tasks
  • Medium - 1500-2500 tokens for moderate tasks
  • Low - 800-1200 tokens for simple queries