/model silently hides Extra High for GPT-5.4 without explaining tier gating in Copilot CLI 1.0.36

[daideguchi]

Describe the bug

Issue #2725 was closed with the explanation that Extra High / xhigh reasoning effort is intentionally restricted to Business and Enterprise plan tiers.

However, in the current CLI this still presents as a confusing and inconsistent experience:

• /model -> GPT-5.4 only shows Low / Medium / High
• the picker gives no explanation that Extra High is being hidden because of plan-tier gating
• current client bundles still appear to contain xhigh-related labels/descriptions/handling paths, which makes this look like a silent UI filter rather than a clearly communicated product restriction

So the problem here is not "please add xhigh back", but rather:

• the restriction is silently applied in the picker
• the CLI does not clearly explain why the option is missing
• this is easy to interpret as a broken or inconsistent model selector

Related issues:

• GPT-5.4 /model picker hides Extra High even though xhigh still works in 1.0.27 #2725
• xhigh reasoning was removed for gpt-5.4 and gpt-5.3-codex! #2739

Affected version

GitHub Copilot CLI 1.0.36

Steps to reproduce the behavior

1. Start copilot
2. Run /model
3. Select GPT-5.4
4. Open the reasoning effort picker
5. Observe that only Low / Medium / High are shown
6. Observe that there is no inline explanation that Extra High is hidden due to plan-tier restrictions

Expected behavior

If Extra High is intentionally unavailable for the current account tier, the CLI should make that explicit.

For example, any of these would be much clearer than silently omitting it:

• show Extra High as disabled with a short explanation
• show a short message in the picker that it is restricted by plan tier
• document the restriction in a place the CLI can point to
• ensure the UI/messaging is consistent with the actual supported-effort metadata exposed by the client

Additional context

Environment:

• Windows
• GitHub Copilot CLI 1.0.36

I am not reporting this as "xhigh definitely works and should be re-enabled".
This report is specifically about the current CLI UX being ambiguous and easy to misinterpret.

After reading #2725, I understand the intended restriction.
The problem is that the current /model picker still hides the option without clearly telling the user that this is a plan-tier gate rather than a broken selector.

[daideguchi]

Adding one potentially related symptom:

In some flows, the selected reasoning effort appears to snap back to medium immediately after model switching / selection, even when a different effort was chosen.

This looks related to:

• effortLevel config is not respected when switching models via /model command #2823
• xhigh reasoning was removed for gpt-5.4 and gpt-5.3-codex! #2739

I am not sure whether this is the exact same root cause as the silent tier-gating / hidden-option problem described above, but from a user perspective both show up as the reasoning-effort UI not matching the effective result.

If they are different issues internally, it may still be helpful to keep them linked.