# Cursor AI Model Configuration Guide

**Last Updated:** 2026-04-04

This guide defines how to configure Cursor for **quality-first delivery** with **token/cost control**.

## Core Policy

1. Pick models by **capability tier**, not hardcoded names.
2. Keep Max Mode **off by default**; use it only when a large coupled context is required.
3. Prefer **Ask/read-only** for discovery and **Agent** only for implementation.
4. Use `@` scoped files, not broad folder attachments, for day-to-day work.
5. **Composer 2** (2026): Cursor positions this family of models for **stronger token efficiency** on coding tasks ([Introducing Composer 2](https://cursor.com/blog/composer-2)). Re-check **Settings → Models** after upgrades so defaults match your quality bar.

Official references (Cursor may serve both `cursor.com/docs` and `docs.cursor.com`; if a link 404s, use in-app **Help → Docs** search):

- Models / context: [cursor.com/docs](https://cursor.com/docs) (navigate to Models / Context)
- Max Mode: [docs.cursor.com/context/max-mode](https://docs.cursor.com/context/max-mode) (or search “Max Mode” in Cursor Help)
- @ Files & folders: [docs.cursor.com/context/%40-symbols/%40-files-and-folders](https://docs.cursor.com/context/%40-symbols/%40-files-and-folders)
- Usage limits: [cursor.com/help/models-and-usage/usage-limits](https://cursor.com/help/models-and-usage/usage-limits)
- Agent modes: [cursor.com/docs/agent/modes](https://cursor.com/docs/agent/modes)

## Recommended Defaults (Ordio)


| Task type                                                | Mode             | Model tier              |
| -------------------------------------------------------- | ---------------- | ----------------------- |
| Exploration, “where is X”, read-only checks              | Ask              | Fast model              |
| Single-file, low-risk edit                               | Chat/Composer    | Fast or balanced model  |
| Multi-file refactor, shared components, ambiguous logic  | Agent            | Premium reasoning model |
| Large coupled refactors needing broad context continuity | Agent + Max Mode | Premium reasoning model |


## Practical Setup

1. Open Cursor Settings → Models.
2. Set default to a **premium reasoning model**.
3. Keep a **fast model** available for low-risk tasks.
4. Enable Auto only if quality remains stable for your workflows.
5. Verify your active model and usage limits in Cursor account/settings.

## Cost & Token Rules

1. Start a new chat for unrelated workstreams (prevents context accumulation).
2. Attach only files you will edit or inspect.
3. Avoid Full Folder Content unless absolutely required.
4. Break broad “do everything” requests into smaller waves.
5. Use local scripts/validators first; avoid repeated MCP/browser loops when local checks suffice.

## Freshness Policy

Model availability and pricing can change quickly. Re-verify this doc against official Cursor docs before major updates to team defaults.