Design passes
Use the model with the
best design judgment
best design judgment
Choosing Models in AI Manager
Best-practice AI Manager model choice for design, coding, and orchestration, with preferred Claude and GPT defaults.
Choose models from the context of the whole AI Manager workflow
Model choice is only one part of the broader AI Manager workflow. Read it in context with the current task thread, the toolbar controls, the provider account state, and the prompt you are about to send. That wider context is what tells you whether the next slice needs design judgment or implementation discipline.
The model dropdown is where you match the engine to the current slice
Once a provider is connected, the next practical choice is the model itself. Treat this dropdown as a task-fitting control, not as a permanent identity badge. The right model for design exploration is often not the same model you want for programming, orchestration, and validation-heavy execution.
Programming passes
Use the model that is strongest
for code and structure
for code and structure
Orchestration
Prefer models that stay strong
across larger task slices
across larger task slices
Verification
Keep review and validation in
the loop whatever you choose
the loop whatever you choose
With the dropdown open, inspect the live models available from this provider
This live menu shows the models the current provider is actually exposing to AI Manager right now. The exact list can vary by provider and account, so use this moment to scan what is available before choosing the model that best fits the current slice.
Provider-dependent
The available list changes
with the connected provider
with the connected provider
Account-dependent
Your account tier can change
which models appear
which models appear
Task-dependent
Pick from the live list for the
slice you are solving now
slice you are solving now
Prefer Claude Sonnet 4.6 and especially Opus 4.7 for design-heavy work
When those models are available in the live list, Claude Sonnet 4.6 and especially Opus 4.7 are strong defaults for design choices. They tend to be better when the task is about hierarchy, layout judgment, UI alternatives, copy tone, and the kind of visual reasoning that makes the result feel intentionally designed instead of merely assembled.
Hierarchy
Strong at page structure
and visual ordering
and visual ordering
UI critique
Useful for comparing layouts,
states, and polish
states, and polish
Tone and direction
Better for copy feel and
design intent
design intent
Prefer GPT-5.4 and GPT-5.5 for programming and orchestration
When available in the same live menu, GPT-5.4 and GPT-5.5 are the better defaults for programming, larger execution slices, and orchestration across files and surfaces. Use them when the job is about implementation discipline, code generation, follow-up fixes, file-to-file consistency, and keeping a broader agentic workflow moving without losing the thread.
Programming
Strong for implementation,
editing, and refactors
editing, and refactors
Cross-file work
Good when the task spans
more than one file
more than one file
Orchestration
Better for planning, execution,
and validation loops
and validation loops
Switch models by slice, not by hype
A practical workflow often starts with a design-oriented model for the interface direction, then switches to a programming-oriented model for implementation and follow-up fixes. That is usually stronger than forcing one model to handle every phase equally well.
Use Claude first when the slice is mostly about design direction, critique, or UI alternatives.
Switch to GPT when the slice becomes implementation, orchestration, or validation heavy.
Judge each pass by the owning Wappler surface, not by model branding alone.
Text
Design-first handoff example
"First, improve the dashboard layout, hierarchy, and empty states. After the visual direction is solid, switch to implementation mode and generate the concrete HTML, App Connect bindings, and follow-up fixes in reviewable slices."
Pick the model that fits the current slice, then keep reviewing in Wappler
Use Claude Sonnet 4.6 or especially Opus 4.7 when design judgment matters most. Use GPT-5.4 or GPT-5.5 when programming and orchestration matter most. Then review the result in the owning Wappler surface and keep the next prompt narrow enough that the model choice stays intentional.
