Skip to main content
Some models can think before they answer: they work through a problem internally, then produce a final reply informed by that work. This improves quality on hard problems, and it costs more tokens and time. This page covers how to get reasoning in the plugin, a worked example, and when it is worth using.

​
Reasoning Is a Property of the Model

In the plugin, reasoning is not a flag you pass. There is no --reasoning option. You get reasoning by choosing a reasoning-capable model. The catalog marks which models support reasoning, so browse it and pick one:
Once you select a reasoning-capable model, it thinks through your prompt on its own. See Choosing a model for how to pick one.

​
A Worked Example

Pick a reasoning-capable model and ask it to reason step by step through a problem where the intuitive answer is wrong:
You can run the same thing from your terminal with nito ask. See Ask and Fusion from the terminal. A reasoning-capable model works through it and lands on the correct answer, **0.05(5cents)∗∗,nottheintuitive10cents.Theintuitiveanswerfailsbecauseiftheballwere10centsthebatwouldbe0.05 (5 cents)**, not the intuitive 10 cents. The intuitive answer fails because if the ball were 10 cents the bat would be 1.10, making the total $1.20. Working it out step by step catches this.

​
When to Use It

Reasoning trades tokens for quality. It costs more and takes longer, so use it for genuinely hard problems (math, multi-step logic, careful analysis) and leave it off for routine generation where it adds cost without improving the answer. When in doubt, browse the catalog with models and match the model to the difficulty of the task. See Choosing a model.

Choosing a Model

Browse the catalog and pick a model that fits the task.

Ask and Continue

Send a prompt and keep the conversation going.

Features Overview

What Nito can do across models and commands.