Subagents start in default model, then transition to subagent model

What version of Codex CLI is running?

codex-cli 0.154.0

What subscription do you have?

Pro

Which model were you using?

gpt-5.6-astra, gpt-5.6-luna

What platform is your computer?

ubuntu server 26.04

What terminal emulator and version are you using (if applicable)?

herdr

Codex doctor report

What issue are you seeing?

I have the following config:

model = "gpt-5.6-sol"
model_reasoning_effort = "xhigh"
personality = "pragmatic"
approvals_reviewer = "auto_review"
plan_mode_reasoning_effort = "xhigh"
service_tier = "default"

[agents]
enabled = true
max_concurrent_threads_per_session = 8

default_subagent_model = "gpt-5.6-luna"
default_subagent_reasoning_effort = "max"

When a subagent is spawned, it is spawned in whatever context the orchestration window is in (astra, sol, etc.). It takes many turns reading AGENTS.md, oritenting itself, etc. Then it switches to the actual subagent model and then passes along all this information yet again.

What steps can reproduce the bug?

use a subagent different from the main orchestrator agent

What is the expected behavior?

subagent starts in the specified model. Currently this is burning a ton of tokens in my orchestrator model and little in the desired subagent model

Additional information

No response

Start by reproducing the issue with the reported [agents] configuration, spawning a subagent with a different model from the orchestrator. Trace subagent model selection during startup; done means default_subagent_model is used from the first subagent turn without initial work in the orchestrator model.