Responses: reasoning tokens spent but Desktop thinking empty when only encrypted_content / empty summary arrives #38160

What happened

On Codex Desktop (and CLI), some Responses turns show:

Codex is reasoning, but the upstream Responses API exposed this reasoning block only as encrypted private reasoning…

or no thinking panel content at all, even when:

• model_reasoning_effort is high/ultra
• model_reasoning_summary is auto or detailed
• usage reports output_tokens_details.reasoning_tokens > 0

So the model did think, but the UI has nothing readable to show.

Why it matters

Users cannot tell whether the agent is making progress during long agentic/Goal turns. Effort sliders appear active, but thinking is blank or a dead-end encrypted notice.

Expected

When model_reasoning_summary is not none and the Responses API returns a reasoning item:

1. Prefer streaming/showing summary text (summary_text / response.reasoning_summary_text.delta).
2. If only encrypted_content is present, either:
   • still request/attach a plaintext summary when the account/model supports it, or
   • document clearly that this backend path cannot surface thinking (not a misconfigured effort slider).

Repro sketch

1. Codex Desktop with model_reasoning_summary = "detailed" and a GPT-5.6-class model.
2. Long agentic turn (tools + Goal).
3. Observe thinking panel: encrypted-only notice or empty, while token usage includes reasoning tokens.

Related config

model_reasoning_effort = "ultra"
model_reasoning_summary = "detailed"

Catalog field default_reasoning_summary defaults to "none" in the official models cache, which makes this easy to hit if the session does not override it.

Environment

• Codex CLI/Desktop recent (0.147.x-class)
• Responses wire API
• No logs/secrets attached

Ask

• Confirm whether empty summary + non-zero reasoning_tokens is expected for some backends.
• If not, fix request/response handling so detailed/auto summaries reliably populate the thinking panel for agentic turns.