OpenAI to Integrate Codex into ChatGPT, Shifting Focus from Chat to Execution

OpenAI's super app has finally taken shape.

The name is still ChatGPT, but internally, a new member—Codex—is about to join.

And ChatGPT is basically just a shell.

At OpenAI's recent "Intelligence at Work" event, the company officially announced that Codex will be integrated into ChatGPT over the coming weeks.

At first, I didn’t quite process this news—wasn’t there always a Codex option in ChatGPT?

I only tried it today and found out it’s fake—it just prompts you to download the Codex app, and you can’t use it directly within ChatGPT.

So, once the merge is complete, you’ll be able to chat and get work done in a single app—no more switching between apps.

But do you think things end here?

From OpenAI’s perspective at least, Codex may be just the beginning.

Regarding all the possibilities of the super app, we finally clarified everything through this launch event.

The press conference lasted a full hour with high information density, but in summary, it boils down to three key points.

1. ChatGPT is merging with Codex because times have changed.

2. Codex will continue to be upgraded, so three major updates have been released.

3. Codex is catching up to Claude Code, with GPT-5.5’s “savings” being a key factor.

Here’s a quick overview for everyone:

First, the merger event that surprised many: why did ChatGPT choose to merge with Codex now?

One is that Codex was initially launched to catch up with Claude Code, and now it has truly proven its results.

Codex's monthly active users have surpassed 5 million, a sixfold increase since the desktop version launched in February.

More importantly, 20% are not developers at all, but knowledge workers such as analysts, designers, and investment banking professionals—and this group is growing three times faster than developers.

This indicates that Codex is gaining strong momentum beyond its existing community.

Moreover, regarding the revenue issue that OpenAI highly values (it initially pursued Claude Code due to its strong monetization potential), OpenAI has revealed that enterprise revenue currently accounts for 40% of its total revenue, and this proportion is expected to reach 50% by the end of this year.

In short, Codex is experiencing rapid user growth, reaching a wide audience, and is genuinely generating revenue for OpenAI.

It seems everything is telling OpenAI that it’s time to consider the next step.

The next step forward is, at least for now, very clear—moving from Chat to Agent, from conversation to execution.

Alexander Embiricos, Head of Products at OpenAI, said at the launch event:

You may not work 24/7, but your agent in the cloud will.

A single line captures the ultimate appeal of a super app: in the future, users won’t need to switch between apps—just give a command, ChatGPT will understand it, and Codex will invoke various agents to execute it.

So this integration is essentially OpenAI's first step toward becoming a super app.

You ask what’s next? The answer is probably something everyone has already guessed—your browser.

The browser is the final piece of this integration, providing AI with an entry point into the web world, allowing users to automatically perform tasks such as searching, operating backends, and handling workflows—tasks that previously required manual clicks—simply by speaking in ChatGPT through Codex.

至此, ChatGPT + Codex + Atlas browser, OpenAI is finally poised to achieve end-to-end execution through a single app.

But today, let’s just talk about Codex.

To give Codex more “soldiers” to get things done, OpenAI has rolled out three more updates at once:

Six core role agents with plugins covering data analysis, sales, creative production, product design, equity investment, and investment banking, directly integrated with 62 enterprise applications such as Snowflake, Figma, and Salesforce, and featuring 110 built-in automation skills.

The Annotations feature allows you to select and edit directly on the original document, eliminating the need to regenerate the entire file. This capability is already in use by developers to modify code, Markdown files, and websites generated by Codex.

Now, it has expanded to content creation scenarios such as documents, spreadsheets, and presentations.

The Sites feature lets Codex turn your work into an interactive web app with one click, generating a URL you can instantly share with your team.

This feature feels quite practical. In the past, non-technical people could only vaguely describe their requirements when developing a project; now, everyone can generate a clear, intuitive visual demo, significantly reducing team communication costs.

This feature is currently available in preview for Business and Enterprise customers.

Did you see that? OpenAI is betting that, in the future, an increasing number of ordinary people will shift from conversation to execution.

Someone might ask, then why didn’t Codex acquire ChatGPT?

It can only be said that, on the surface, this appears to be the story of ChatGPT consuming Codex, but when several details are put together, the direction may be exactly the opposite.

In the organizational restructuring in May, Thibault Sottiaux, who led the growth of Codex, was promoted to Head of Core Products and Platform, overseeing all three lines: consumer, enterprise, and developer. Meanwhile, Nick Turley, who drove ChatGPT to 900 million weekly active users, was reassigned to the enterprise product line.

So now it’s Codex’s people managing ChatGPT, not the other way around.

Additionally, the wording of Brockman’s internal memo is worth considering—

He didn't write "upgrade ChatGPT," but rather "invest in a unified agentic platform."

Therefore, this merger is less about product integration and more about a strategic shift: from conversation to execution, from chat to Agent.

ChatGPT provides a user base of one billion, while Codex provides a growth engine and a narrative for the future.

The shell is ChatGPT's, but the soul is Codex's.

However, Codex's existence was ultimately forced into being by its competitor, Anthropic.

Even GPT-5.5, mentioned during the launch event, can almost be considered a product of this pressure.

Looking back, without Anthropic, Codex might still be just an obscure auxiliary feature within ChatGPT today.

Let’s go back to last year when the Claude Code preview was released.

Claude Code achieved immediate success, with its annualized revenue surpassing $2.5 billion in February this year—reaching $1 billion in just six months since launch, making it one of the fastest-growing products in commercial software history.

Approximately 4% of global public GitHub submissions originate from it, with users averaging 20 hours of engagement per week.

It was precisely the surge of Claude Code that made OpenAI realize it had fallen behind in the field of coding.

What if you're falling behind? Then catch up.

Previously, the default narrative in the AI industry was that Anthropic was following OpenAI’s lead—Claude would follow after GPT, and Claude would position itself as a counterpart to ChatGPT.

But on the Code front, the script has been completely rewritten.

Claude Code launched in February last year, and OpenAI didn’t release its counterpart, Codex, until two months later; Anthropic first developed a desktop version, and OpenAI followed suit; Anthropic introduced Cowork for knowledge workers, and OpenAI later caught up with Plugins.

This time, OpenAI is feeling its way across the river, following Anthropic.

However, the pace of catching up is indeed rapid. Over a timeline of 14 months, it grew from zero to 5 million weekly active users.

In April last year, the Codex Cloud preview launched;

In February this year, launched the desktop app and the dedicated model GPT-5.3-Codex;

In March, the super app announced its merger and acquisition of the Python tools provider Astral;

In April, launch a $100/month ChatGPT Pro offer to attract developers;

Mobile app in May;

Officially integrated with ChatGPT in June.

You could say that Claude Code's influence is present in nearly every step here.

To be honest, Codex didn't turn things around by directly competing on code quality.

After all, the blind test data speaks for itself: Claude Code still has a 67% win rate, and most programmers acknowledge that the code it produces is more robust.

But unfortunately, Claude Code’s usage limits are extremely restrictive, and its price is higher than Codex’s.

A consensus distilled from a discussion thread on Reddit involving over 500 developers:

Claude Code is smarter, but its limits are too restrictive for daily use; Codex is slightly less capable but actually works.

A smarter model that frequently gets stuck halfway through tasks and costs more, versus a slightly less intelligent but more affordable model.

It's you— which one would you pick? (doge)

Interestingly, at the launch event, OpenAI’s product lead Alexander also said this memorable quote:

The one thing GPT-5.5 hates most is wasting tokens.

Using Codex with GPT-5.5, you achieve the same quality output with only one-third the tokens.

In short, on the path to catching up with Claude Code, Codex is treating “fewer tokens, more intelligence” as its new guiding principle.

Undeniably, the AI programming field is currently dominated by a two-player rivalry.

One represents the strongest coding ability, and the other represents the strongest productization ability.

But if we look at it over a longer time frame, it becomes clear that Claude Code and Codex are no longer competing just for the programming tools market.

As more ordinary people begin to treat agents as daily work partners, the entry point is truly the most important thing.

This, in turn, creates new opportunities for Chinese players—because in the Agent era, victory is not determined solely by model capability, but also by scenario mastery, ecosystem integration, application connectivity, and understanding of local enterprises’ workflows.

In fact, major domestic companies have gradually entered the space.

Among them, some chose to access the developer portal, others the enterprise internal office portal, and still others the end-to-end intelligent agent platform.

But regardless of the path taken, the goal is essentially the same—to secure the core entry point of the Agent era.

Everyone understands that the real priority is leveraging the time window of local scenarios to build genuine competitive advantages at the Agent infrastructure level.

So, who will be the first to emerge as China’s version of Claude Code and Codex?

And more crucially, who will become the new super entry point in China in the Agent era?

Returning to this merger event.

Given Codex's current trajectory and the broader industry environment, I’d like to boldly say:

Rather than continuing to call the merged app ChatGPT, it would be better to call it Codex directly.

Think carefully—ChatGPT sounds like a relic from the “Chat” era, but OpenAI’s current focus is no longer on chat.

The product is managed by Codex; the internal memo refers to the agentic platform, with the fastest-growing use cases being execution scenarios, and none of the new features are related to chat.

Can ChatGPT be used? Yes. But is it accurate? No.

Of course, OpenAI is unlikely to change its name. After all, "ChatGPT" has become synonymous with AI and represents a brand asset of incalculable value.

But not changing the name doesn't mean the essence hasn't changed.

Soon, when you open ChatGPT, you may no longer be greeted by a chatbox waiting for your question, but by an agent that has already completed your tasks.

By then, whether or not to chat won't matter anymore.

In other words, ChatGPT may become a symbol of spirit in the future…

Reference link:

[1]https://www.techmeme.com/260602/p14#a260602p14

[2]https://openai.com/index/codex-for-every-role-tool-workflow/

[3] https://www.theinformation.com/articles/inside-openais-decision-combine-codex-chatgpt

This article is from the WeChat public account "Quantum Bit," authored by: Focused on Frontier Technologies