Codex

How to Use Another GPT Model in Codex

Open config.toml and find:

model = "gpt-5.5"

Change it to the model you want to enable, then save the file.

After saving, restart the client you are using for the change to take effect: Codex CLI, Codex App, or the editor extension.

Important: after enabling it, do not switch models inside the same long-running conversation. If you already switched models, restart Codex with an explicit command.

CLI Usage

Start Codex with:

codex -m gpt-5.5 -c model_reasoning_effort="xhigh"

If codex is already open, press Ctrl + C to exit, then restart it with the command above.

Codex Still Calls the Official Endpoint

Check your environment:

echo $OPENAI_BASE_URL
echo $OPENAI_API_KEY

Windows PowerShell:

$env:OPENAI_BASE_URL
$env:OPENAI_API_KEY

The base URL should be:

https://api.fhrouter.com/v1

Model Not Found

Use one of the model ids returned by:

curl https://api.fhrouter.com/v1/models \
  -H "Authorization: Bearer YOUR_FHROUTER_API_KEY"

Common GPT/Codex ids:

gpt-5.4-mini
gpt-5.4
gpt-5.5

Windows Command Problems

If codex fails in PowerShell, try:

codex.cmd

If output is garbled:

chcp 65001

Better Prompts for Coding Tasks

Codex works best when the task is scoped. Instead of asking it to build a whole system at once, split the task into smaller steps:

inspect the current code path
explain the likely change
edit one module
run the relevant tests
summarize the diff

Start with read-only work before allowing edits:

codex exec "Explain the project structure without changing files."

Context Gets Too Large

If Codex starts losing focus in a long session, start a new session and provide the current task, relevant file paths, and the latest error. Very long sessions can become harder to control.