When to specialize the model.

"Just use the best model" is not a strategy. Cost, latency, context windows, image tools, and data-handling rules all change which stack wins for a job.

Start agnostic, then specialize

• *Illumora's Studio defaults to an *agnostic
• *target: prompts that travel. Specialize when one of these is true:

• *You need a vendor-specific capability (Claude's long-context drafting, OpenAI structured outputs, Gemini multimodal).
• *You are optimizing for cost or speed on a repeated workflow.
• *Privacy or residency rules constrain which providers you can use.
• *The job is image generation or heavy code agents with tool use.

How to choose

Use the Model Atlas for capability and pricing facts. Use Model Match when you have a job description and want a scored recommendation. Deep-link back into the Studio with ?model= when you are ready to generate.

A practical rule

If you cannot name the specialization reason in one sentence, stay agnostic. Premature vendor lock-in is how prompts rot when the next model drops.