The setup guide shows how to configure Codex CLI with the CheapClankers API base URL, your key and a compatible model ID. Compare the available GPT models and rates below before choosing one.
This configuration uses prepaid CheapClankers API credit. It does not sell OpenAI Codex accounts, ChatGPT subscriptions or an official Codex plan. Tool and protocol support depends on the selected model.
Configure Codex CLI
Install Codex CLI using the official installation instructions. This setup uses prepaid developer API credit, separately from any ChatGPT or Codex subscription.
1. Set your key in the environment
export CLANKER_API_KEY="your_key_here"In PowerShell use $env:CLANKER_API_KEY = "your_key_here". Keep the key out of source control and launch Codex from the same shell.
2. Configure the Responses provider
Create a CheapClankers API key, export it as CLANKER_API_KEY, then add this provider to ~/.codex/config.toml. Replace the model ID with an online model that supports the Responses API and the tools your workflow needs.
model_provider = "cheapclankers"
model = "YOUR_ONLINE_RESPONSES_MODEL"
[model_providers.cheapclankers]
name = "CheapClankers"
base_url = "https://api.cheapclankers.com/v1"
env_key = "CLANKER_API_KEY"
wire_api = "responses"Merge these settings into your user configuration; avoid duplicate TOML keys or replacing unrelated settings. An available GPT model is not necessarily Responses-compatible. Check the model details and your key’s access before running Codex.
3. Check a small request
Run codex and ask it to reply with hello without editing files or running commands. Expect a short reply and a corresponding entry in dashboard usage history. Responses support alone does not guarantee all agent tools work; verify your required workflow before a larger task.
Troubleshooting
A 401 can mean a missing or invalid key; check that the environment is visible to the shell launching Codex. A 403 can indicate restricted key access. A 404 can mean the model or Responses protocol is unavailable. Check wallet credit for insufficient-credit errors, and check usage before retrying interrupted requests.
Review the official custom-provider configuration when upgrading Codex. For other clients, follow the Node.js SDK, Python SDK or Claude Code tutorial.
Compare credit packs · Read API setup docs · Developer setup tutorials · Open the setup guide
Models, availability and token rates
Gpt 5.6 Terra
Model verified through the auto group.
Gpt 5.6 Sol Openai Compact
Model verified through the auto group.
Gpt 5.6 Sol
Model verified through the auto group.
Gpt 6.1 Terra
Online in the auto group. Usage is estimated where published rates are available and reconciled against supplier records.
Gpt 6 Astra
Online in the auto group. Usage is estimated where published rates are available and reconciled against supplier records.
Gpt 6 Sol
Online in the auto group. Usage is estimated where published rates are available and reconciled against supplier records.
Gpt 6.1 Sol
Model verified through the auto group.
Gpt 5.6 Luna
Model verified through the auto group.
Gpt 6.2 Sol
Online in the auto group. Usage is estimated where published rates are available and reconciled against supplier records.
Gpt 6.2 Astra
Online in the auto group. Usage is estimated where published rates are available and reconciled against supplier records.
Prices are USD usage credit per one million tokens. Cached input can have a different rate. Consult model details before use.
Frequently asked questions
Is this a cheap Codex subscription or account?
No. It is a third-party API endpoint that can be configured in Codex CLI for supported models. Official Codex and ChatGPT subscriptions are separate.
How do Codex CLI requests use credit?
Requests to a configured model consume prepaid USD usage credit at that model's published token rates. The exact amount depends on input, output and any cached tokens.
Which model should I select for Codex CLI?
Use an online model whose protocol and tool support fit your workflow. The setup guide supplies a configuration example; test the chosen model before relying on it for a project.