Make Codex usage last longer

OpenAI says model choice, context, reasoning, tool use and caching all affect how much of your allowance a task uses, and cloud tasks may use more than local ones. These settings and commands come from the Codex docs. Each tip below cites the official doc it comes from (read 2026-10-20). We don't estimate how much each one saves; vendors don't publish that.

1. Check usage and context

   /status shows the active model and remaining context. /usage shows daily, weekly or cumulative token activity.

   /status
   /usage weekly

   learn.chatgpt.com checked 2026-10-20

2. Leave Fast mode off unless you need it

   Fast mode uses included subscription limits at 2.5x the Standard rate (2x for purchased credits). Ultrafast uses included limits at 8x. /fast toggles it; to keep it off by default, do not set service_tier = "fast".

   /fast        # toggles; run again to turn off
   
   # ~/.codex/config.toml: leave this line out
   # service_tier = "fast"

   learn.chatgpt.com checked 2026-10-20

3. Pick the model and reasoning effort

   /model sets the model and, where available, the reasoning effort. Persist a default in config.toml. Levels depend on the model.

   /model
   
   # ~/.codex/config.toml
   model_reasoning_effort = "medium"

   learn.chatgpt.com checked 2026-10-20

4. Start fresh between tasks

   /new starts a new chat in the same CLI session; /clear also clears the terminal.

   /new
   /clear

   learn.chatgpt.com checked 2026-10-20

5. Compact long runs

   /compact summarises the chat to free tokens. You can set when auto-compaction triggers (unset = model default).

   /compact
   
   # ~/.codex/config.toml (pick your own value)
   # model_auto_compact_token_limit = <tokens>

   learn.chatgpt.com checked 2026-10-20

6. Keep AGENTS.md lean

   AGENTS.md is read into project instructions. project_doc_max_bytes caps how much is read.

   # ~/.codex/config.toml (pick your own value)
   # project_doc_max_bytes = <bytes>

   learn.chatgpt.com checked 2026-10-20

7. Prefer local over cloud for small tasks

   OpenAI: cloud tasks may use more of your allowance than local messages.

   codex   # run locally in your repo

   learn.chatgpt.com checked 2026-10-20