# Cursor Auto mode vs Copilot auto vs AutoRouter

AutoRouter · Oct 23, 2026 · 10 min read

Cursor's Auto mode and GitHub Copilot's auto model selection each pick a model for every request, inside their own tools. Copilot shows which model answered each response; Cursor hides the routed model by default unless a team admin turns it on. AutoRouter, the Mac app for Codex and Claude Code, picks a model for each turn in those two apps, shows which model handled it, what it cost and why, and lets you steer it with plain-English rules or the app's model picker.

AutoRouter wrote this post. What it says about Cursor, GitHub Copilot, Codex and Claude Code comes from their own documentation, read on October 22, 2026, and links to it.

## How does Cursor's Auto mode pick a model?

Cursor's Auto picks the model for each request within a mode you choose, and you can't hand-pick the model while Auto is on. With Cursor Router on, it runs a classifier on each agent request and routes it, by task type and complexity, to the most cost-effective model that still gives comparable quality, according to Cursor's router docs. The modes:

• - Cost is classic Auto. It uses the previous Auto routing logic and optimizes token spend.
• - Balance optimizes for intelligence, speed and cost.
• - Intelligence sends harder tasks to the most capable models.

Balance and Intelligence need Cursor Router, which is available only on Teams and Enterprise plans, and they use your limits faster than Cost. As of October 22, 2026, individual plans have Cost mode only, and the $8-a-month Plus plan has no Auto. Cursor's docs say Cost bills from your weekly pool whatever model serves it, and Balance and Intelligence bill at the routed model's list price; Cursor's help center says every Auto mode bills at list price, so check your own plan. On Teams and Enterprise, a Balance or Intelligence route to an eligible third-party model adds Cursor's Token Rate of $0.25 per million tokens (as of October 22, 2026).

As of October 22, 2026, Cursor's help center lists GPT-5.5, Claude Opus 5, Grok 4.6 and Claude Fable 5 in the router's pool and says the pool changes over time. Cursor hides the routed model by default and says that lets you judge results on merit; team admins can show it at the start of each response in Balance and Intelligence.

## How does GitHub Copilot's auto model selection pick a model?

Copilot's auto combines one system that tracks each model's real-time health and availability with another that judges how complex the task is, and routes each task to a model your plan and your admins' policies allow, per GitHub's docs. It can change models at natural cache boundaries and when a conversation's complexity shifts a lot.

In VS Code, Copilot CLI and the GitHub Copilot app you can pick a tier: Efficiency puts cost first, Balance weighs cost, quality and latency, and Intelligence puts quality first. The same models are available in every tier; the tier changes which ones it prefers. GitHub doesn't list the models auto chooses from, and on plans for individuals it may serve evaluation models, which you can turn off.

You see the model behind every response: hover over it in Copilot Chat or the Copilot app, read it in the terminal in Copilot CLI, or find it at the end of the response in cloud agent. Usage is charged by the model auto picks, and as of October 22, 2026, paid plans get 10% off model costs while using auto in Copilot Chat, Copilot CLI, the Copilot app and cloud agent. In VS Code, auto can also route to an on-device model you've enabled, and those responses use no AI credits.

## Do Codex and Claude Code have an Auto mode of their own?

No. As of October 22, 2026, neither app's model documentation describes a setting that picks a model for each turn based on the task.

• - Codex uses a recommended model unless you set one with /model or in config.toml. OpenAI's Codex models page suggests GPT-6.1 Sol for complex coding and agentic work and GPT-6 Luna for focused, repeatable tasks, and you make the choice.
• - Claude Code has opusplan, which uses Opus in plan mode and switches to Sonnet for execution, per its model configuration docs. Planning always runs on Opus and execution on Sonnet, whatever the task. Without it, the default setting runs Opus 5.5 on Claude's Pro, Max, Team and Enterprise plans. Its fallback models step in only on errors or safety flags.

## How does AutoRouter pick a model in Codex and Claude Code?

AutoRouter works out how hard each turn is and sends it to a model that clears the quality bar for it: hard work goes to the strongest model your plans cover, and quick tasks go to the cheapest capable model, often your plan or a low-cost open model. Your own rules come first. In an illustration on AutoRouter's landing page, a change across six files goes to Claude Opus 5.5 on the Claude plan, and a commit message goes to a low-cost open model. How AutoRouter picks a model for each turn gives the full order.

AutoRouter adds 20+ models to Codex and Claude Code, plus one more choice, AutoRouter, which picks the model for each turn. You choose it from the model menu in Codex or with /model in Claude Code. In Codex, OpenAI models run on your ChatGPT plan; in Claude Code, Claude models run on your Claude plan.

Every turn shows its model, what it cost and why. You steer it three ways:

• - Rules in plain English, which come before AutoRouter's recommended ones. Examples from its site: "Use Claude Opus 5.5 for UI work." "Prefer GPT-6.1 Sol for code reviews." You can see how often each rule is used.
• - The model picker. Choose a specific model in Codex or Claude Code and your turns go straight to it. AutoRouter routes only while it is the model selected.
• - Ratings. If a pick falls short, rate it. The AutoRouter team reviews real routing decisions and every rating people send, and retunes the router with them.

## How do Cursor Auto, Copilot auto and AutoRouter compare?

They differ most in where they run, whether you see the pick and its cost, and how you steer it. Details are as of October 22, 2026.

## Which one should you use?

Use the Auto built into the tool you already work in. Each of the three routes only inside its own tools.

• - You work in Cursor: use Cursor's Auto. It's built in, and on Teams and Enterprise an admin can set its modes for the whole team. AutoRouter works only with Codex and Claude Code for now.
• - You use Copilot in VS Code, JetBrains or on GitHub: use Copilot's auto. It shows the model for every response.
• - You only want Opus to plan and Sonnet to build in Claude Code: opusplan does that with nothing to install.
• - You work in Codex or Claude Code on a Mac and want a pick on every turn that you can see and change: use AutoRouter. It's in beta, sometimes misjudges how hard a task is (you can rate the pick), and doesn't publish benchmarks yet.
• - You run a team on Codex or Claude Code: AutoRouter Enterprise is priced by seats and rolls out to every Mac through Jamf or another MDM; see how to roll out AI coding tools to a team.

## Questions

### Is Auto mode good enough for hard tasks?

Each of the three can send hard tasks to a capable model, and each lets you switch to a specific model when a pick falls short. Cursor recommends Intelligence for complex, multi-step work, and Copilot's Intelligence tier puts quality first. AutoRouter sends hard work to the strongest model your plans cover and never treats unclear work as a quick task; Does model routing make your results worse? goes into the trade-off.

### Does Auto mode cost less than choosing a model yourself?

It can, and each vendor describes the saving its own way. As of October 22, 2026, Copilot takes 10% off model costs on paid plans when you use auto, and Cursor says its Intelligence mode costs less than running a single frontier model. AutoRouter's landing page gives an example week, which it presents as an illustration: with 30% hard, 45% medium and 25% quick turns at Claude Opus, Sonnet and Haiku 5.5 list prices, AutoRouter's Free plan does 1.9 times the work of using a frontier model for everything, and AutoRouter Pro does 2.9 times.

### How does AutoRouter make money?

AutoRouter earns the 10% it adds to the provider's price on paid models, and it earns nothing on your ChatGPT or Claude plans. Price only decides between models that already clear the quality bar, and your plans come first. Pro costs $20 a month and includes $20 of usage each month; paid models bill from that usage first and then from credits you top up.

### What does AutoRouter keep from each turn?

AutoRouter keeps the message that opens each turn, up to 4,000 characters, to review routing, along with records of the model chosen, tokens and cost. Requests pass through AutoRouter's servers on their way to each provider, and AutoRouter doesn't store transcripts. Details are in AutoRouter's privacy policy.

### Does AutoRouter work outside Codex and Claude Code?

Not yet. AutoRouter works with Codex (the app and the CLI) and the Claude Code CLI on macOS 14 or later, and Claude Code needs Node 22.18 or later. More apps are on the way.

## Sources

• - AutoRouter
• - AutoRouter privacy policy
• - Cursor Docs: Cursor Router, read October 22, 2026
• - Cursor Docs: Models and pricing, read October 22, 2026
• - Cursor Help: Available models, read October 22, 2026
• - Cursor Help: Cursor Router, read October 22, 2026
• - GitHub Docs: Auto model selection, read October 22, 2026
• - VS Code Docs: Language models, page dated October 21, 2026, read October 22, 2026
• - OpenAI: Codex models, read October 22, 2026
• - Claude Code Docs: Model configuration, read October 22, 2026