You can now call GPT-5.6-sol inside Claude Code lol
Moreover, overseas it's already a hot topic that "GPT-5.6-sol is clearly better inside Claude Code than inside Codex..."
OpenAI's latest model running in Anthropic's tool.
And it performs better than the original.
It doesn't make sense, right? lol
I'll explain it step by step.
But before that,
Since AI is evolving so fast that many people can't keep up with the information,
I've created an open chat where I teach everything from the latest AI info to monetization methods 👇
https://xytgf.notion.site/38c4361b791b809f819dc93555b5f468?source=copy_link
■ Is Sol better in Claude Code than in CodeX!?
Why does it perform better in Claude Code than in the original (CodeX)?
This is the main point: what separated the performance wasn't the model itself, but the harness (foundation).
Think of the model as the engine and Claude Code as the car body.
Even with the same engine, if the car body design is better, it runs faster.
Claude Code has a thick build for the "outside of the model," such as how it calls tools, controls sub-agents, and brings out the depth of thought.
A practitioner followed by hundreds of thousands in the developer community, the creator of the T3 stack, also says:
"Claude Code is way ahead in that you can fix the subagent's effort yourself."
If the contents are the same, the one with the more careful exterior shows more strength.
That's all it is.
■ So, how do you actually do it? (2 routes you can try today)
So, specifically, what should you do?
I'll leave it here in a form you can copy and paste today. It's the exact procedure I save.
Just one premise.
If you are paying for ChatGPT's paid plan (Codex tier),
With just that right, you can run OpenAI's latest coding model "GPT-5.6 Sol" inside Claude Code.
No separate API billing is required.
The mechanism is like putting an interpreter in between. Claude Code can only speak Anthropic-style language.
Sol only replies in OpenAI style.
Since they can't talk directly, you sandwich an interpreter called CLIProxyAPI (a proxy that runs on your PC).
Use your ChatGPT paid login as the interpreter's ID.
That's why no additional charges occur.
Route A = Official Plugin Version (Safe, used alongside Claude)
1/plugin marketplace add openai/codex-plugin-cc2/plugin install codex@openai-codex3/reload-plugins4/codex:setup
Now you can use /codex:review, /codex:adversarial-review, and /codex:rescue.
Claude writes, GPT critiques, and you ship.
A way to add GPT as a reviewer while keeping Claude as the main character.
This is safer if you're trying it for the first time.
Route B = Proxy Version (Using Sol as the main)
Step 1: Install CLIProxyAPI (Just write port number 8317 and an API key you decided on in the config.yaml file)
Step 2: Connect (Authenticate through ChatGPT and Claude, then connect to Claude Code)
Step 3: Define the following alias and enjoy claudex
The alias in Step 3 is divided into two depending on your usage preference.
Route B-1 = Sol Always Main Version
1alias claudex='ANTHROPIC_BASE_URL=http://127.0.0.1:8317 \2ANTHROPIC_AUTH_TOKEN=your_key \3CLAUDE_CODE_SUBAGENT_MODEL=gpt-5.6-sol \4CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 \5CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY=3 \6ENABLE_TOOL_SEARCH=false \7claude --model gpt-5.6-sol'
Sol is the main from the moment you type claudex.
You can also fix the subagent model and effort (depth of thought) with environment variables.
For those who want to use Sol seriously.
Route B-2 = Model Switching Type (Usually Claude, Sol only when needed)
1alias claudex='ANTHROPIC_BASE_URL=http://127.0.0.1:8317 \2ANTHROPIC_AUTH_TOKEN=your_key \3claude'
This stays as the usual Claude,
and only switches to Sol when you type /model gpt-5.6-sol.
Switching back is also just one /model command.
The difference is essentially just whether you add --model gpt-5.6-sol at the end.
I use this one.
Because you can keep Claude's decision-making and only throw second opinions or heavy implementation requests to Sol.
To summarize, the differences between the three are:
▼ Route A (GPT as a reviewer)
- Start with
claudeas usual - Claude writes, GPT critiques with
/codex:review - Requirements: Plugin + Codex CLI -> For those who want to try it safely first
▼ Route B-1 (Sol as always main)
- It's Sol from the moment you start with
claudex - All responses become Sol
- Requirements: CLIProxyAPI + 2 authentications -> For those who want to use Sol seriously
▼ Route B-2 (Usually Claude, sometimes Sol)
- Start with
claudex, usually the usual Claude - Sol only when you type
/model gpt-5.6-sol - Requirements: Same as B-1 (just remove --model) -> For those who want to keep Claude's decision-making (I'm this one)
If you throw this article into Claude Code, the procedure itself will be finished in a few minutes.
■ What is this all about, anyway?
The answer to "Which AI model should I use?" is no longer about choosing one.
It's about deciding on the vessel (harness) first and then how to arrange who inside it.
Use Claude Code as the foundation, with Fable as the orchestrator and Sol, Grok, and Sonnet as the task force.
When a new model comes out, don't switch everything; just add a new force to the task force.
This Sol commotion was also part of this orchestration story when you get down to it.
■ How to survive the AI era?
Finally, I'll talk about something a bit bigger.
AI is evolving so fast that
"Another new model?"
"Which one should I use in the end?"
"How should I master it?"
I think many people worry about these things.
The day ends just by researching, and before you know it, nothing has taken shape.
Because many of my followers are studious, I want to dare to say...
If you only follow information, you'll stay a "person who knows" for the rest of your life.
The real meaning of this story is
"Individuals can now have an 'AI team'."
Have Fable orchestrate and Sol or Sonnet do the work.
This is what companies used to do by hiring several people not long ago.
The designer, the implementer, the reviewer.
Now you can organize an AI team in an instant.
In fact, I've seen many times where work that an outsourcer said "will take a month" finished in 3 hours with Claude Code.
Since the execution cost has almost disappeared, ideas become businesses as they are.
Without hiring people, and with almost zero initial cost.
Thinking calmly, it's an incredible era, isn't it?
And I can say for sure
This is still the "early stage."
Meaningful updates come every 2 to 4 months, and common sense from six months ago no longer applies.
If you compare it to the internet,
it's still like the atmosphere of 1995.
That period when just being able to make a homepage became a job.
In other words, everyone who touches it from now on is on the "early side."
So, the way to survive the AI era from now on is simple.
- Learn how to use the tools (like this article)
- Think about "where to use" them
Unfortunately, 90% of people end up just learning "how to use the tools."
But the difference is overwhelmingly in the latter.
Where you use it is 100 times more important than how you use it.
And to be honest, I'm a little angry.
Even though such tools are open to everyone,
ordinary people are being shut out from the entrance just because of the atmosphere of "it looks difficult" or "it's for engineers."
In the meantime, only a few people with fast information are quietly making money.
I have decided to change this structure from the ground up.
That's why I follow overseas primary information every day, actually run it in my own environment,
and transmit the procedures that I was able to reproduce like this.
Even if you can't write code, don't have time, or don't have business knowledge,
it doesn't matter anymore.
I'll step on the difficult parts and check them first,
so please use your saved time to think about "where to use" it.
And I will teach not only the latest AI information
but also how to launch a business and monetize using Claude!
AI the wave to reduce the number of people left behind by the AI wave by even one.
I'm posting this with serious intent, not as an exaggeration.
So please don't just finish reading it.
Let's go to the early adopter side together while it's still the "early stage."
Whether you're on the "side that started then" or the "side that just watched" a few years from now.
The fork in the road is a day like today.
If you found this article helpful, jump to my profile
and don't forget to follow and turn on notifications!
In the next article,
I will specifically write about the contents of that organization, what to leave to the orchestrator, and how to arrange the task force.
