---
title: "One run, the right model at every step"
description: "Ara routes each step to the model that fits the work — coding, reasoning, review, speed, or cost. Bring Claude, Codex, local models, or your own keys."
url: "https://ara.so/product/model-routing"
section: "Platform"
---

# One run, the right model at every step

Ara routes each step to the model that fits the work — coding, reasoning, review, speed, or cost. Bring Claude, Codex, local models, or your own keys.

## Match the model to the work

### Code generation

Heavy refactors and edits land on the strongest coding model in your roster.

### Reasoning + review

Planning and diff review go to a model tuned for careful reasoning.

### Speed and cost

Lint fixes, renames, and chores route to a fast, cheap model — or a local one.

## One run, many models

### Classify the task

Ara reads the signal and splits the run into steps — code, reasoning, review, chore. Each step gets a model class, not one global pick.

_Triage_

### Route to the best fit

Each step goes to the model that scores best on your policy — quality, latency, or price. Claude, Codex, local, or your own key.

_Dispatch_

### Verify, then escalate

Cheap models take the first pass. If checks fail or confidence drops, Ara escalates to a stronger model and retries before opening the pull request.

_Pull request_

## Route every run to the right model

Connect your keys, set a policy, and let Ara pick the model for each step — then ship the verified pull request.