Skip to main content

Use GPT Models in Claude Code

4i.codes8/13/26About 3 min

CC-Switch can convert requests from Claude Code to the OpenAI Responses API format and forward them to GPT models through its local routing service. After setup, you still start Claude Code with the claude command and use /model to view or switch the mapped models.

Before You Start

Make sure that:

  • The latest version of CC-Switch is installed and running.
  • Claude Code is installed and the claude command works.
  • You have created an API key for a Codex service group in the 4i.codes console. See API Keys and Access for key creation details.

Important

You must use a key from a Codex service group, not a Claude service group. Treat the API key as a credential: never share it or commit it to a public repository.

Step 1: Add a Claude Code Provider

Open CC-Switch, select Claude Code in the top navigation, and click the + button in the upper-right corner to add a provider.

Make sure Claude Code is selected rather than Codex. Although this provider ultimately calls a GPT model, it receives requests from Claude Code and therefore belongs in the Claude Code provider list.

Select Claude Code and add a provider in CC-Switch
Select Claude Code and add a provider in CC-Switch

Step 2: Configure the Provider and Model Mapping

Enter the following values on the provider page:

FieldValueNotes
Provider name4icode-gptYou can use another recognizable name
Websitehttps://4i.codesOptional provider reference
API KeyA Codex service-group keyCreate and obtain it from the key list in the 4i.codes console
API endpointhttps://api.4i.codesDo not append /v1 or another path
API formatOpenAI Responses API (Routing Required)This option requires CC-Switch routing to be enabled later
Authentication fieldANTHROPIC_AUTH_TOKEN (Default)Keep the default value

Warning

The API format is essential. Do not select the Anthropic API or OpenAI Chat Completions format. Select OpenAI Responses API (Routing Required).

Configure Model Mapping next:

  1. Click Fetch Model List in the upper-right corner of the model-mapping section.
  2. For each role, select the GPT model to use under “Actual Request Model.”
  3. Map Sonnet, Opus, Fable, Haiku, and the default fallback to available GPT models. If you use Subagent, map it to a GPT model as well. To use one model consistently, map all of them to gpt-5.6-sol.
  4. The display name appears in Claude Code's /model menu. It can be customized, but using the actual model name makes the mapping easier to recognize.
  5. Review the configuration and click Save in the lower-right corner.

Tips

Claude Code sends different types of requests using role names such as Sonnet, Opus, and Haiku. Model mapping determines which GPT model handles each role. The default fallback handles requests that do not match a specific role.

Enter provider details and map GPT models
Enter provider details and map GPT models

Step 3: Open CC-Switch Settings

Return to the CC-Switch home screen and click the Settings icon next to the CC-Switch logo in the upper-left corner.

Open CC-Switch settings
Open CC-Switch settings

Step 4: Enable Claude Routing

Open the Routing tab in Settings, then:

  1. Turn on the Routing Master Switch.
  2. Turn on Claude in the routing applications section.
  3. Confirm that the local routing service shows a running status.
  4. Save the settings and leave the Settings page.

Both the master switch and the Claude switch must be enabled. CC-Switch must also remain running so that it can forward Claude Code requests through the local routing service.

Enable the routing master switch and Claude routing
Enable the routing master switch and Claude routing

Step 5: Verify the Model in Claude Code

If Claude Code is already open, exit it completely and start it again:

claude

Then enter:

/model

The model menu displays the names configured in Step 2. Select a GPT model and send a simple test message. A successful response confirms that CC-Switch is routing Claude Code requests to the selected GPT model.

View mapped models with the model command in Claude Code
View mapped models with the model command in Claude Code

Troubleshooting

The mapped GPT models do not appear in /model

  • Confirm that the provider was added under Claude Code, not Codex.
  • Confirm that both the display names and actual request models were saved in Model Mapping.
  • If you use Subagent, confirm that it is also mapped to an available GPT model.
  • Exit Claude Code completely and run claude again.

Requests are not reaching a GPT model

  • Confirm that the API format is OpenAI Responses API (Routing Required).
  • Confirm that both the routing master switch and Claude routing are enabled.
  • Keep CC-Switch running and verify that the local routing service shows a running status.

Authentication, model-not-found, or endpoint errors

  • Use a valid Codex service-group key and confirm that the account has sufficient balance and access.
  • Set the endpoint to exactly https://api.4i.codes without /v1 at the end.
  • Fetch the model list again and select a GPT model that is currently available.

Return to the original Claude models

Open Settings → Routing in CC-Switch and turn off Claude routing or the routing master switch, then restart Claude Code. To use another Claude provider, return to the Claude Code provider list and activate that configuration.

4i.codes