Reasoning model
Published Updated
What is Reasoning model?
A reasoning model is an AI model that works through an internal chain of thought, planning and checking its steps, before giving a final answer. It trades speed and cost for accuracy on hard problems.
A specific type of AI model that generates internal chain-of-thought tokens to plan, verify, and self-correct before outputting a final response.
Reasoning used to come in separate models such as OpenAI's o1 and DeepSeek-R1. Claude and GPT models now expose that tradeoff through adjustable reasoning effort, letting you choose how much deliberation to use for a task.
How it relates to Orbit
Orbit lets a complexity tier choose a model and effort level for a coding agent. See Complexity Tiers & Routing for details.
When you would use it
You use a reasoning model for tasks where getting the exact correct answer is critical and requires deep logic, accepting that the model will take longer to respond than a standard chat model.
Common operations
- Resolving complex architectural bugs.
- Generating highly accurate code for advanced algorithms.
Frequently asked questions
When is a reasoning model worth the cost?
For multi-step logic, debugging and architecture work. Routine edits rarely need one.
Can I set reasoning effort per prompt in Orbit?
Yes. Each tier has an effort level, and you can override it for a single prompt. See Tune Effort Per Prompt.