iOS Remote realtime Voice has no model selector, uses the wrong configured voice, and cannot resume

Summary

ChatGPT Remote on iOS exposes Codex realtime Voice as a separate, one-shot flow rather than as a capability of a normal new Codex chat:

• A normal new Codex chat created in iOS Remote has no realtime Voice control.
• Realtime Voice can only be started through the separate New Realtime Voice Chat action.
• That iOS voice flow has no conversation-model selector.
• After ending Voice, the same chat cannot start Voice again; another new realtime voice chat must be created.
• The configured Sol voice is not honored on iOS Remote: a different male-sounding voice is played.
• When the iOS-created realtime voice chat is opened on macOS, its conversation-model selector is also unavailable.

By contrast, a voice chat created directly on macOS allows selecting GPT-5.6 Sol, uses the configured Sol voice correctly, and can restart Voice in the same chat after ending.

Environment

• macOS ChatGPT app: 26.721.41059
• ChatGPT for iOS: latest App Store version available as of July 28, 2026 (no update offered)
• Surface: Codex through Remote on iOS
• Configured account voice: Sol

Expected behavior

At minimum, the iOS Remote voice flow should:

1. Allow realtime Voice to be started or resumed in a normal Codex chat, or clearly preserve the separate voice chat as a resumable session.
2. Expose a conversation-model selector, or inherit and display the intended conversation model.
3. Honor the account's configured Sol voice.
4. Preserve model metadata so the iOS-created chat exposes the same model controls when opened on macOS.
5. Allow Voice to restart in the same chat after ending.

Actual behavior

1. Creating a normal new Codex chat in iOS Remote does not provide realtime Voice.
2. The only way to use realtime Voice is to select the separate New Realtime Voice Chat action.
3. This iOS realtime voice flow does not expose any conversation-model selector.
4. Opening that iOS-created voice chat on macOS still does not expose a conversation-model selector.
5. Although the account voice is configured as Sol, iOS Remote plays a different male-sounding voice. The configured Sol voice works correctly in a voice chat started directly on macOS.
6. After Voice is ended on iOS, the same chat cannot start Voice again. The user must create another new realtime voice chat.
7. A macOS-created voice chat behaves differently: it allows selecting GPT-5.6 Sol, uses the configured Sol voice, and can restart Voice after ending.

Reproduction

1. In ChatGPT settings, configure the voice as Sol.
2. Pair ChatGPT for iOS with a macOS ChatGPT/Codex host using Remote.
3. On iOS Remote, create a normal new Codex chat.
4. Observe that the normal chat has no realtime Voice control.
5. Return and select the separate New Realtime Voice Chat action.
6. Observe that the voice chat starts without any conversation-model selector.
7. Observe that a different male-sounding voice is used instead of the configured Sol voice.
8. End Voice.
9. Reopen the same chat on iOS and observe that Voice cannot be restarted.
10. Open the same iOS-created voice chat on macOS and observe that its conversation-model selector is unavailable.
11. Create a new voice chat directly on macOS and observe the control behavior: GPT-5.6 Sol can be selected, the configured Sol voice is used, and Voice can be restarted after ending.

Additional information

This looks like an iOS Remote voice-session initialization and state-continuity problem involving:

• the distinction between normal Codex chats and the separate realtime voice flow,
• missing conversation-model metadata and controls,
• failure to apply the configured account voice,
• and failure to preserve a resumable voice session.

The issue has also been submitted through the in-app /feedback flow with diagnostics. Screenshots are available and can be added if needed.