Question: Model selection for inline subagents when using copilot engine

[ryckmansm]

Hi team,

I’m trying to better understand how model selection works for inline subagents when using the copilot engine in gh-aw.

Context

From my understanding, workflows can define a model (or use a default), and (inline) subagents can also define a model to use.
This is also mentioned in the documentation: https://github.github.com/gh-aw/reference/inline-sub-agents/

My observation

Based on what I observe, it seems that:

Inline subagents always use the model defined at the workflow level, regardless of any intended variation at the subagent level. I only use copilot as engine, so I don't know if the observation holds for other engines.

In other words, I don’t see evidence that subagents are actually switching models.

What I’m trying to validate

Could you confirm whether this observation is correct?

• Is this expected behavior?
• If not, could this indicate a bug?
• Or is my understanding of how model selection works flawed?

Observation from gh-aw itself

I tried to validate this using the existing “large vs small model” experiment in the repo, but I don’t see any difference in the logs that would indicate model switching at the subagent level:

• Large model run:
  https://github.com/github/gh-aw/actions/runs/27908514467

• Small model run:
  https://github.com/github/gh-aw/actions/runs/27892685844

Despite the different configurations, the logs look effectively identical from a model usage perspective (at least based on what is exposed).

Key question

Where can I explicitly verify which model is being used by inline subagents?

• Is there a way to surface this in logs?
• Is there a debug flag or telemetry that shows per-agent model usage?
• Or is the model resolved strictly at the workflow level and not per subagent?

Why this matters

I’m exploring scenarios where different subagents would use different model sizes (e.g., cost vs. reasoning trade-offs). Understanding how model resolution actually works is important for designing these workflows correctly.

---

Thanks in advance for clarifying — and for pointing me to the relevant code or documentation if I’ve misunderstood the behavior.

[ryckmansm]

I think the daily documentation report agent misunderstood my question. I know that the subagent defaults to the parent workflow model when the subagent doesn't define a model.
What I'm seeing is that, when the subagent has defined a model (different than the parent workflow model) the subagent still uses the parent workflow's model.
Or at least that is what I think I see happening.