[FEATURE] Subagents having agnostic models based on depth and complexity

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

I would like for subagents to have an agnostic model-choice where the agent or session that spawns the subagent takes an evaluation and decides the effort level and choice of the subagents model.

This is nice to have when you have agents for debugging in increasing depth and complexity of tasks. The model can then be chosen depending on how difficult the task is at hand, while still having one singular bug-fixer subagent

Theres also smaller tasks and bugs that would require smaller models, and certain models (such as Fable) which can only be used for certain usage, therefore not redeeming the Subagent unavailable if you have used all your Fable usage.

If you already manage to get claude to create the subagents on its own, it already takes this evaluation on the model choice and effort-level so I just want this dynamically. So dynamic sub-agents

Proposed Solution

When creating a subagent you have the model specification be either ANY, where claude can then choose the model itself. Or you can have it be Opus|Fable where the Models then gets a choice of those 2 models

This can easily be done by Claude understanding the task, and before picking the model evaluating the complexity and which model it has the need for. You can also make it favor certain models if you are worried about usage.

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

Developer tools/SDK

Use Case Example

Example Scenario:

1. I am going in for a bug fix and Fable spawns a bug-hunter subagent.
2. The subagent had an agnostic model. Therefore Fable takes an evaluated decision based on its knowledge around the models and their capabilities, and the difficulty of the task which model can be used here.

Additional Context

_No response_