How to change reasoning effort in Codex CLI: model_reasoning_effort values and one-off overrides

A developer has documented how to control reasoning effort in OpenAI's Codex CLI through the model_reasoning_effort configuration key, which accepts five levels: minimal, low, medium, high, and xhigh. The guide covers setting the value permanently in ~/.codex/config.toml, overriding it for a single run with the -c flag, switching mid-session via the /model command, and bundling settings into reusable profiles. It also notes that plan_mode_reasoning_effort and agents.default_subagent_reasoning_effort apply effort settings only to plan mode or spawned subagents, and warns that values unsupported by the active model may not take effect.

Originally published at https://aicoding-guide.com https://aicoding-guide.com/en/posts/codex-reasoning-effort/ . Codex feels slow, or it overthinks a trivial fix. The knob for that is reasoning effort , the Codex counterpart of extended thinking in Claude Code, controlled by the model reasoning effort key. In short: set model reasoning effort = "high" or another value in ~/.codex/config.toml to change it permanently, use the -c flag for a single run, or /model during a session. The five levels are minimal , low , medium , high , and xhigh . Key point What you will learn - The five values and when each fits - Four ways to change effort: permanently, for one run, mid-session, and per profile - The keys that change effort only for plan mode or subagents The configuration reference defines model reasoning effort as a string with the values minimal | low | medium | high | xhigh . The reference does not state a default, and it may depend on the model, so none is given here. | Value | Suited to | |---|---| | minimal | one-line questions, formatting fixes, anything with little to reason about | | low | small bug fixes, simple refactors | | medium | everyday feature work and investigation | | high | multi-file design changes, bugs with no obvious cause | | xhigh | hard algorithms, reviews of large changes where mistakes are costly | Higher effort means slower responses and more tokens. Keeping everyday work around medium and raising it only for hard tasks is the easiest routine. Glossary Reasoning effort : how much the model reasons internally before answering. It is separate from how much of the reasoning summary is displayed; hiding the display is covered in Hiding reasoning in Codex output https://aicoding-guide.com/en/posts/codex-hide-agent-reasoning/ . One line in ~/.codex/config.toml : model = "gpt-5.5" model reasoning effort = "high" In a trusted project you can put it in the repository's .codex/config.toml to affect only that project; project settings outrank user settings. To change effort for a single invocation without touching the file, use -c --config . Values are parsed as TOML , so wrap the string in double quotes and wrap the whole thing in single quotes to keep the shell from eating them. codex -c model reasoning effort='"high"' "Find out why this test fails and fix it" The same form works with codex exec from scripts. According to the docs, a value that cannot be parsed as TOML is treated as a string, so a missing pair of quotes often still works, but quoting makes the intent explicit. Run /model in a session to switch. The official command list describes it as choosing the active model "and reasoning effort, when available". Raise it before a hard investigation and drop it afterwards. Rather than typing the override every time you review code, bundle the settings in a profile. Profiles live in $CODEX HOME default ~/.codex as <name .config.toml and are selected with --profile <name . ~/.codex/deep-review.config.toml model reasoning effort = "xhigh" approval policy = "on-request" codex --profile deep-review Keep only the differences in the profile and shared settings in the base config.toml . Profiles as a whole are covered in the hub article Configuring Codex with config.toml https://aicoding-guide.com/en/posts/codex-config-toml/ . The configuration reference also lists keys that apply only in specific situations. | Key | Applies to | |---|---| | plan mode reasoning effort | overrides effort in plan mode only | | agents.default subagent reasoning effort | default effort for spawned subagents | If you want deep planning but ordinary implementation, add plan mode reasoning effort = "high" and leave model reasoning effort alone. If nothing changes, check the model A value outside what the model supports may not take effect. If /model shows no reasoning effort choices for the current model, that model does not expose effort. When speed and quality do not change after editing the setting, confirm the model first. model reasoning effort , with values minimal , low , medium , high , xhigh ~/.codex/config.toml ; one run: -c model reasoning effort='"high"' ; mid-session: /model <name .config.toml profile and select it with --profile plan mode reasoning effort ; subagents have agents.default subagent reasoning effort