Choosing a model
Every helper has a default model — the AI it uses when you run it. Margaret gives you three options, all from Anthropic's Claude family. They differ in how deeply they think, how long they take, and what they cost.
The short version: Sonnet 4.6 is the right choice for most helpers. Start there, and only change it if you have a reason.
The three models
Haiku 4.5 — Fast and light
Haiku is the quickest and cheapest option. It's well suited to tasks where the instructions are clear and the output is relatively structured — things like formatting, simple summaries, extracting key points from short text, or filling in a template.
It's not the best choice for tasks that require judgement, nuance, or handling ambiguous inputs. If your helper needs to weigh things up or make considered recommendations, Haiku may produce results that feel thin.
Good for: Formatting tasks, simple summaries, templated outputs, high-volume tasks where cost matters.
Sonnet 4.6 — The reliable workhorse
Sonnet is where most helpers should live. It handles structured analysis, well-defined workflows, research summaries, and multi-section outputs reliably. It's significantly more capable than Haiku for anything involving reasoning or judgement, and noticeably faster and cheaper than Opus.
If you're not sure which model to use, pick Sonnet.
Good for: Most things — client communications, analysis, proposals, meeting notes, research summaries.
Opus 4.7 — Deep thinking
Opus is the most capable model in the family. It's built for tasks that are genuinely complex — where the input is ambiguous, the stakes are high, or the output requires careful, layered reasoning. It takes a bit longer and costs more per run, but for the right tasks the quality difference is real.
Use it when Sonnet is giving you results that feel incomplete or when the task genuinely requires careful, considered judgement.
Good for: Complex strategic analysis, long document review, tasks where nuance really matters, situations where you'd want a very senior person looking at it.
How to change the model
Each helper stores its own default model. You can change it from the helper panel — look for the Model section below the helper name. Changes take effect from the next run.
You can also ask Margaret to recommend a model for a specific helper. Click Recommend in the Model section and Margaret will read the helper's instructions and suggest the best fit, with a reason.
A note on cost
Opus costs roughly 5× more per run than Sonnet, and Sonnet costs roughly 3× more than Haiku. For most individual runs the amounts are still small (see Understanding costs), but if you're running a helper many times a day it's worth thinking about.
The dashboard shows you a cost breakdown by helper so you can see where your spend is going.
Which model is being used right now?
The Use button in the helper panel always shows which model will be used: "Use with Sonnet 4.6" etc. If no model has been saved for a helper, it defaults to Sonnet 4.6.