Skip to main content

GPT models in Claude Code

Do not use GPT models directly in Claude Code

Claude Code is built for Anthropic-style requests. GPT and Codex models are OpenAI-compatible models.

For normal use, keep them separate:

  • Claude models in Claude Code.
  • GPT/Codex models in Codex CLI or OpenAI-compatible clients.
  • Gemini models in Gemini CLI or Gemini-compatible clients.

Claude Code expects Anthropic-style request and response behavior. GPT/Codex models are OpenAI-compatible. Forcing GPT/Codex models through Claude Code usually requires a local compatibility bridge, and that bridge can break tool calls, caching, streaming, model selection, or usage accounting.

What you should use instead

For GPT/Codex models, use Codex CLI with:

https://api.fhrouter.com/v1

Example models:

gpt-5.4-mini
gpt-5.4
gpt-5.5

Advanced experimental setup

If you already maintain your own local compatibility bridge, configure that bridge to call FHRouter as an OpenAI-compatible API:

Base URL: https://api.fhrouter.com/v1
API key: your FHRouter token key
Model: gpt-5.4-mini

Then point Claude Code at the local bridge endpoint, not directly at https://api.fhrouter.com/v1.

At that point, the FHRouter side of the flow is only the key, endpoint, and model path. Bridge-specific behavior remains part of your own local setup.

Copyright (c) 2026 FHRouter
Copyright © 2026 FHRouter