experimental feature "collab" is broken on main for API/free/plus/go users

What version of Codex CLI is running?

main

What subscription do you have?

N/A

Which model were you using?

gpt-5.2-codex

What platform is your computer?

_No response_

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

N/A

What issue are you seeing?

After @jif-oai merged #11745, hardcoding the Explorer AgentRole to have a fixed model override of gpt-5.3-codex-spark where it used to be gpt-5.1-codex-mini, trying to use the Experimental collab feature no longer works for users on a Free or Plus subscription as they don't have access to Codex Spark, as well as people using Codex CLI via API.

What steps can reproduce the bug?

Enable collab via /experimental or via editing ~/.codex/config.toml, and watch as it fails to explore due to model unavailability.

› Hey, use the explorer subagents to explain what models are used for exploring.


• Agent spawned
  └ call: call_9DPDLA2CS27lsprBo65AphhM
    agent: 019c5844-deb4-78e3-aa34-cffbb4be15e7
    status: pending init
    prompt: Find and explain which models are used for explorer subagents in this repo or docs. Look for any configuration/docs about explorer model selection.

• Waiting for agents
  └ call: call_ZR2b8AqVEiBggFUhdilh1PC1
    receivers: 019c5844-deb4-78e3-aa34-cffbb4be15e7

• Wait complete
  └ call: call_ZR2b8AqVEiBggFUhdilh1PC1
    agents: 1 total · 1 errored
    019c5844-deb4-78e3-aa34-cffbb4be15e7 errored: { "error": { "message": "The requested model 'gpt-5.3-codex-spark' does not exist.", "type": "invalid_request_error", "param": "model", "code": "model_not_fo...

What is the expected behavior?

I'd expect that if something is exposed via /experimental it wouldn't regress in such a way.

I'd completely understand this happening (but would still be bummed) if this was marked as Stage.UnderDevelopment.

Additional information

_No response_