Feature Request: One-time Model Addressing in Claude Code Prompts (e.g. `@opus`)
Problem
Currently, when using Claude Code with a default model (e.g. Sonnet), switching to a different model such as Opus requires changing the model setting. While this works, it introduces friction in workflows where only a single task requires the higher-tier model. Afterward, the user must manually switch back to the default model - a step that is easy to forget and may cause unnecessary compute usage or unintended behavior.
Proposed Solution
Introduce inline, one-time model addressing within prompts, similar to how @agents can be invoked. For example:
@opus Can you deeply review the concurrency handling in this class?
In this example, the request is sent to the Opus model, but the session remains on the default (Sonnet) for subsequent prompts. This would give users flexibility to escalate specific tasks without altering the session-wide model setting.
This issue has 5 comments on GitHub. Read the full discussion on GitHub ↗