Steerability

Steerability describes how responsive a model is to direction — how well it follows custom instructions, adapts to a defined persona, changes tone on request, or adjusts the level of detail it provides. A highly steerable model is flexible: a builder can configure it with a system prompt and reliably get consistent, on-spec behavior. A less steerable model may have strong default tendencies that resist redirection — it might revert to its trained defaults even when explicitly instructed to behave differently. For behavior architects, steerability is a key factor when evaluating which model to build on: a capable but poorly steerable model may require much more work to fit a specific product context than a slightly less capable but more responsive one.