Make Claude Fable 5 Use GPT 5.6 Sol: Here's How
In this walkthrough I show you how to combine Claude Fable 5 and GPT 5.6 Sol inside Claude Code, using the official OpenAI Codex plugin. The result is a setup where Fable 5 acts as the director of your business and GPT 5.6 Sol does the actual execution, which also preserves your hard-earned Fable 5 tokens.
What you need
- Claude Code up and running.
- A ChatGPT subscription for the Codex side. A twenty dollar subscription from OpenAI is good enough for this use case.
- The plugin's GitHub link, which I walk through below. The plugin itself is completely free.
1. Combine Claude Fable 5 with GPT 5.6 Sol (0:00)
This plugin was released back in March 2026. It gives Claude Code access to Codex. Codex is the equivalent of Claude Code from OpenAI and runs on the latest GPT models. And to demystify where it lives: GitHub is pretty much just a Google Drive used by developers that lets them keep track of software versions.
2. How the official Codex plugin works (0:11)
It is an official OpenAI plugin that plugs Codex into Claude Code, so you can run Codex-powered code reviews and handoffs to Codex. Grab the link:
https://github.com/openai/codex-plugin-cc
3. Install the plugin in Claude Code (0:44)
Go into Claude Code, paste the link, and tell Claude Code to verify what is in that link and set it up. This is the exact request I used:
Hey Claude, can you please go to that link just using WebFetch and tell me what's inside of there? Then I will need your help setting it up.
4. Why WebFetch is faster than browser automation (1:23)
The reason I said WebFetch: when you ask Claude to get something from the web, it can sometimes use browser automation, and that is super slow. If you know you just need the information inside a webpage, tell it to use fetch. That is how you make sure you do not waste tokens on slow browser automation.
5. What the Codex plugin gives Claude Code (1:48)
Claude browsed the page and confirmed it: a Codex plugin for Claude Code, official from OpenAI, for Codex-powered reviews and handoffs. And to be clear, this is not only for people who code. This is fantastic for any person out there, and the next two sections explain why.
6. Use Claude Fable 5 as the orchestrator (2:07)
Fable 5 is a model that works beautifully as an orchestrator. That means you really do not want to spend your hard-earned Fable 5 tokens doing the work itself. You want to spend them orchestrating all of the workers.
7. Use GPT 5.6 Sol as the execution layer (2:26)
On the other side, GPT 5.6 is extremely good at execution. So you get the best of both worlds: Fable 5 as the director of your business, GPT 5.6 Sol as the executor doing the actual labor.
8. Install and configure the plugin (2:44)
In my case, Claude noticed my workspace already referenced the plugin because I had installed it before, so I asked for a fresh install anyway to show you the flow. The installation takes a couple of seconds. In your case, it will simply install it the first time you ask.
9. Test Claude with two Codex subagents (3:31)
With the plugin installed, I set the model to Fable 5 and ran a real test:
Hey, can you please use two Codex subagents to go ahead and fetch me the latest news in France and also in Spain for AI trends? Make sure that it's very quick, I don't want something very detailed.
Claude Code launched two Codex subagents: one for France, one for Spain, each running on GPT 5.6. You can check everything they are doing from Claude Code itself. GPT 5.6 is now working for you directly inside the Claude Code session.
10. Combine Claude and Codex strengths (4:53)
Why bother combining them? Fable has taste; it is fantastic at creating UI. Codex is extremely powerful at pinpointing security risks in what you are building in the back, and in my opinion it is the better long-term worker. Manage to combine these two and it works out great for you.
11. Save Claude Fable 5 tokens (5:20)
Here is what happens behind the scenes: Claude Code calls two different agents that use Codex. Those agents do not consume your Fable 5 tokens. They use their own tokens from your ChatGPT subscription. A twenty dollar ChatGPT subscription goes very, very far here, and even the one hundred dollar tier gives you a lot more usage than the one hundred dollar tier from Claude. The agents send back their responses, Claude Code hands me the final answer, and I preserved all the Fable 5 tokens I needed by routing the work to something more token efficient.
12. Final recommendations (6:24)
Grab a twenty dollar OpenAI subscription; it is good enough for this use case. And beyond the token savings, having Codex review what Fable or Opus produce is bound to surface errors from time to time. These LLMs behave like, quote-unquote, biased human beings: sometimes they do not see their own mistakes, but they can see each other's.
Pitfalls and tips
- Always say "use WebFetch" when you only need the contents of a page. Browser automation burns tokens and time.
- Do not let the orchestrator do the labor. Fable 5 directs, GPT 5.6 Sol executes.
- Cross-review is the hidden win: models catch each other's mistakes far better than their own.
Where to go next
- Want GPT 5.6 Sol as the main brain of Claude Code instead of a subagent? Read Run Claude Code with GPT 5.6 Sol in 5 Minutes.
- To have Claude find more capabilities like this on its own, see How to Actually Find Any Claude Skill.
