Using more than one model is Smith's signature capability. This is one of the 11 Playbook categories. Each play below is a single move you can run right now, tagged by effort. Filter to the quick ones if you are just getting started.
What is a play?
A play is a single move you can run in Smith right now: one prompt, one result. Each is tagged by effort so you can pick the quick wins first. Plays are grouped into 11 categories; this article is one of them. For multi-step workflows that combine several moves, see the Recipes.
Before you run these plays
Every play here sends one prompt to more than one model, and multi-model is off by default. Turn it on first, otherwise you get a single answer.
You compare the models your project has activated. A set of base models is included with every project, and the line-up changes as new releases land, including newer GPT and Claude generations, so check the Settings panel for what your project can use right now. Advanced models are purchased and then activated, and an advanced model that has not been activated shows in a locked state and cannot be selected by project members, so if a model you expected is missing, ask an admin to activate it. Two models still give you a useful comparison. Each model carries a cost indicator, so you can see the relative cost before you choose.
The Plays
Each card is one play. The badge shows the effort: Quick is a single prompt, Few steps adds a little setup, and Setup is a build-once move that pays off repeatedly. If a response is running long or heading the wrong way, you can cancel it while it is still generating.
Do: Select two or three models, run one prompt, compare the answers.
Get: A decision backed by independent takes, not one opinion
Multi-modelDo: Send the same writing prompt to each model, keep the best lines from each.
Get: A draft better than any single model produced
Multi-modelDo: Ask two models the same factual question; where they disagree, verify.
Get: A built-in error check on anything that matters
Multi-modelDo: Use a fast base model for quick passes, then switch to an advanced reasoning model for the final. Which ones you can pick depends on what your project has activated.
Get: Fast where it's fine, careful where it counts
Multi-modelDo: When two models disagree, ask a third to adjudicate.
Get: A resolution instead of a coin flip
Multi-modelSee It in Action
Here is one of these plays running, so you know what to expect before you try it yourself.
The three-perspective decision
Ask the same question of two or three models and compare, what the glossary calls a multi-model move.
Next Steps: Pick one play and run it on something real today. When single moves feel natural, the Recipes show how to chain them into repeatable workflows. Working through your first month? The First 30 Days plan tells you which plays to use each week.
Related Articles
Playbook 9: Workspace-Grounded Power Moves
The next category of single-move plays.
Playbook 7: Analysis & Data
The previous category of plays.
Smith Recipes
Ready to combine plays into multi-step workflows?
Comments
0 comments
Article is closed for comments.