Cheatsheet

Modelor Auto?

Stage 1 · AdoptionChoosing the tool
Auto for the routine. A deliberate choice for the heavy set. Copilot lets you pick the model or leave the choice to Auto, and most people never touch the setting. Model names change every few months. The decision doesn't. That's why this page doesn't name a single model.

Before you open the model picker

Surface first, model second

Chat for a quick question and answer. Researcher when you need several sources, citations and a report. Cowork when the task has several steps and a defined result.

Sometimes the tool decides

Some tasks come with their model built in: images, or the agents inside your apps. There's nothing to choose. Decide where you actually have a choice.

Leave it on Auto when …

Routine, not risk

Quick questions, mail threads, meeting follow-ups, summaries. Auto matches the model and the effort to each request. Here, what Copilot finds in your mail, chats and files matters more than which model answers.

Writing and rewriting

Auto as well. Tone is easier to fix in your custom instructions than by switching models.

Choose deliberately when …

The heavy set

A paper for leadership, a contract clause, numbers others decide on. Choose the model yourself, turn the effort up and review the result. A better model reduces mistakes. It doesn't replace your review.

Workflow, not one-off

Scheduled prompts, skills, agents, templates your team relies on. Where you can, fix the model. A different model can answer the same prompt differently, so test before you switch.

Prompt

„Here are two answers to the same task. For each: what's correct, what's missing, what would I have to fix before sending it? Then tell me which one you would hand over, and why."

The four-question test

Before the heavy set

Run it when the result matters. It takes ten seconds.

Rules

„1. Is it routine? 2. What does a mistake cost? 3. Will it run again next week? 4. Is the model approved in your organisation? Routine and cheap to fix: Auto. Expensive or recurring: choose deliberately, from what's approved."

Pro moves

Effort is the second dial

More thorough means slower and, where usage is billed, more expensive. Turn it down for routine, up for the deck that matters.

One drafts, another reviews

For research that matters, let one model draft and a different one review. That's the multi-model approach, and where your surface offers it, it's one click.

Cheaper and better are two questions

The model that costs less isn't automatically the weaker one, and the expensive one isn't automatically the right one. Rankings shift with every release. Judge by your task, not by the brand.

Model missing? Ask IT first

Not every model is switched on in every organisation, and each additional provider is a data protection decision. A missing model is governance, not a bug.

The picker gives access. Knowing when to use it is training.

Access isn't fitness.

FAQ

When should I leave Copilot on Auto?

For routine work: quick questions, mail, meeting follow-ups, summaries, writing and rewriting. Auto matches the model and the effort to each request, and for this kind of work what Copilot finds in your mail, chats and files matters more than which model answers. Pascal Brunner-Nikolla's rule: Auto for the routine. A deliberate choice for the heavy set.

When should I choose the model myself?

When a mistake is expensive or the work runs again: a paper for leadership, a contract clause, numbers others decide on, a prompt or agent your team relies on every week. Choose deliberately, turn the effort up for heavy work and review the result yourself. For recurring work, fix the model where you can and test before you switch.

Why doesn't this cheatsheet name any models?

Because model names and versions change every few months, and a cheatsheet full of yesterday's names helps nobody. The rules on this page hold whatever the picker shows. Model names, rankings and costs belong in the newsletter, where they carry a date.

Is a more expensive model always better?

No. Cost and quality are two separate questions, and rankings shift with every release. Higher effort also makes any model slower and, where usage is billed, more expensive. Run the same prompt on two models with your own task and decide from the results.

Why can't I see every model in Copilot?

Your organisation decides which models and providers are switched on, and each additional provider is a data protection decision. Some models stay off until an admin enables them. If a model is missing, ask your IT team: that's governance, not a bug.

Last updated: 9 November 2026

Part of the guideWhen to Use What

More cheatsheets on this topic · Choosing the tool

Every Monday at 7:30 CET, one tool you can put to work the same day.

Subscribe to the newsletter