Hide or expand reasoning in Codex output: hide_agent_reasoning and model_reasoning_summary
The three Codex CLI settings that control reasoning display: hide_agent_reasoning, show_raw_agent_reasoning, and model_reasoning_summary, and how they differ from reasoning effort.
Codex shows summaries of what it is thinking while it works. That is useful for following along, but noise when you are skimming a log or piping codex exec output through a script. Other times you want more detail about why it made a decision.
In short: display is controlled by hide_agent_reasoning (suppress) and show_raw_agent_reasoning (show raw reasoning), and summary granularity by model_reasoning_summary. All three are separate from model_reasoning_effort, which sets how deeply the model reasons.
KEY POINT
What you will learn
- The role and default of the three display keys
- Display settings versus depth (
model_reasoning_effort) - Cases where the settings have no effect because of the model or provider
The three keys
hide_agent_reasoning = true # suppress reasoning events
show_raw_agent_reasoning = false # surface raw reasoning when the model emits it
model_reasoning_summary = "concise" # auto | concise | detailed | none| Key | Type | Default | Meaning |
|---|---|---|---|
hide_agent_reasoning | Boolean | false | true suppresses reasoning events in the output |
show_raw_agent_reasoning | Boolean | unset | show raw reasoning content when the model returns it |
model_reasoning_summary | string | unset | summary granularity: auto / concise / detailed / none |
With hide_agent_reasoning = true, reasoning events disappear and only tool activity and the final response remain. model_reasoning_summary = "none" stops summaries being generated at all, which is a different layer from merely hiding them.
用語解説
Reasoning summary: a short, display-oriented digest of the model's internal reasoning. Whether the raw reasoning itself is returned depends on the model and provider; if it is not, show_raw_agent_reasoning = true shows nothing.
Display is not depth
How deeply the model reasons is set by model_reasoning_effort: minimal, low, medium, high, or xhigh. Hiding the display does not reduce reasoning or the tokens it consumes.
| Goal | Setting |
|---|---|
| hide reasoning from the output | hide_agent_reasoning = true |
| reason less and respond faster | model_reasoning_effort = "low" |
| see more of the rationale | model_reasoning_summary = "detailed" |
| see raw reasoning (supported models only) | show_raw_agent_reasoning = true |
Choosing a reasoning effort is covered in the hub article Configuring Codex with config.toml.
Non-interactive use
When you drive Codex from scripts with codex exec, reasoning summaries mixed into stdout make post-processing awkward. A CI profile with hide_agent_reasoning = true keeps the output stable. Profiles live in separate files such as $CODEX_HOME/ci.config.toml and are selected with codex --profile ci.
# ~/.codex/ci.config.toml
hide_agent_reasoning = true
model_reasoning_summary = "none"
approval_policy = "never"Behavior depends on the model and provider
The docs state that show_raw_agent_reasoning has no effect with models or providers that do not emit raw reasoning. model_supports_reasoning_summaries can force reasoning metadata to be sent, but the display does not change if the model does not support it. When a setting appears to do nothing, check the model first.
Another output-readability setting, opening file references in your editor, is covered in file_opener, and turn-completion notifications in notify and tui.notifications.
Summary
hide_agent_reasoning = truehides reasoning;show_raw_agent_reasoning = trueshows raw reasoning when availablemodel_reasoning_summarysets granularity:auto,concise,detailed,none- Depth is a different key,
model_reasoning_effort; hiding does not reduce reasoning or tokens - A
codex execprofile with reasoning hidden gives stable output - Raw reasoning appears only when the model and provider support it
FAQ
- Does hiding reasoning stop the model from reasoning?
- No. hide_agent_reasoning only suppresses the display. Reasoning depth is set separately with model_reasoning_effort.
- show_raw_agent_reasoning = true changes nothing.
- Some models and providers do not emit raw reasoning, and the docs note the setting has no effect in those cases.
- I just want shorter summaries.
- Set model_reasoning_summary = "concise". none disables summaries entirely.
Primary sources
This article was drafted by AI from official documentation and reviewed by the site operator before publishing. Found a mistake? Let us know via the contact page.