OpenAI CEO Sam Altman has declared a “code red” effort to improve ChatGPT, according to an internal memo obtained by The Information. The move represents a dramatic role reversal from three years ago, when Google declared its own “code red” in response to ChatGPT‘s launch.
“We are at a critical time for ChatGPT.”
As one observer perfectly framed it: “It’s almost like Star Wars; with Google, the empire has struck back. Now it all comes down to this: will we see Return of the Jedi, and can OpenAI beat Google again?”
What “Code Red” Actually Means
Altman is pausing ads, AI agents, and Pulse to reallocate people and compute into fixing and upgrading ChatGPT as competition from Google and others intensifies.
| Delayed | Accelerated |
|---|---|
| Advertising experiments | New reasoning model (next week) |
| AI shopping agents | Imagegen improvements |
| Healthcare AI tools | Personalization features |
| Pulse personal assistant | Speed & reliability upgrades |
Altman told staff that ChatGPT needs work on:
- Better personalization for 800 million+ weekly users
- Faster and more reliable responses
- Fewer unnecessary refusals (the “over-refusal” problem)
- Improved “model behavior” so users prefer ChatGPT in public leaderboards like LMArena
He’s encouraging temporary team transfers and instituted daily calls for everyone working on ChatGPT fixes.
The New Reasoning Model: Beating Gemini 3
According to The Information, OpenAI plans to ship a new reasoning model as early as next week that Altman says scores ahead of Google’s Gemini 3 in internal evaluations.
This model will power features like:
- Thinking mode: Extended reasoning for complex problems
- Deep Research: Multi-step investigation and synthesis
- Advanced problem-solving: Math, coding, logic puzzles
As one commenter insightfully noted: “If the new reasoning model really outperforms Gemini 3 internally, we’re entering a new phase: it’s no longer about ‘the biggest model,’ but the most useful, personalized, and less frustrating. This shift matters more than any product launch.”
But Is It Just “Benchmaxing”?
One skeptical user raised a critical concern: “I fear this is benchmaxing over and over, and neglecting real world improvements.”
This captures a real tension: models can score high on standardized tests while still frustrating users with robotic responses or basic errors. OpenAI’s November update to GPT-5 addressed this after users complained it was “too robotic and struggled with simple math problems and geography facts.”
The Google Threat: Gemini’s Rapid Rise
Google’s Gemini has exploded in growth:
| Month | Monthly Active Users | Growth |
|---|---|---|
| July 2025 | 450 million | — |
| October 2025 | 650 million | +44% |
Gemini 3, which debuted November 18, now leads benchmark leaderboards for text generation, image editing, and text-to-image conversion — putting it ahead of ChatGPT, xAI’s Grok, and Anthropic’s Claude.
Over one million users tried Gemini 3 in its first 24 hours. Salesforce CEO Marc Benioff declared on X: “I’m not going back to ChatGPT after trying Google’s new model. The leap is insane—reasoning, speed, images, video… everything is sharper and faster.”
The Imagegen Arms Race
Image generation is now a named priority after Google’s Nano Banana Pro launched to strong reviews. OpenAI wants ChatGPT’s Imagegen to stay competitive for creative and commercial use cases.
This shift acknowledges that visual AI is no longer a “nice to have” — it’s central to user retention, especially as Midjourney, DALL-E competitors, and now Google’s integrated image tools raise expectations.
The Strategic Stakes: Revenue and Funding
ChatGPT currently handles ~70% of global “assistant activity” and ~10% of search, according to Nick Turley, OpenAI’s head of ChatGPT.
The Revenue Projections
| Year | Projected Revenue |
|---|---|
| 2025 | $10 billion |
| 2026 | $20 billion |
| 2027 | ~$35 billion |
OpenAI is counting on this growth trajectory to justify raising roughly $100 billion more amid mounting pressure from Google and Anthropic.
But as Cryptopolitan notes: “Here’s the thing about OpenAI: it’s not making money. The company has to keep raising funds just to stay alive.” OpenAI has committed to spend hundreds of billions on data centers, and investors are getting nervous about when that investment will actually pay off.
User Reactions: Fear, Realism, and Perspective
“As a student, it’s terrifying”
One student captured the existential anxiety: “The architecture I study today is legacy code by next week.”
Another responded with hard-won wisdom: “Oh gosh, yes. But even back in my day, most of what I was taught in computing was out of date when I got into work. I did nearly all my learning at work. Getting work experience is what I suggest to all students. That’s what we look for over anything else.”
“Product experience > monetization (for now)”
One commenter praised the strategic shift: “If the new reasoning model really outperforms Gemini 3 internally, we’re entering a new phase: it’s no longer about ‘the biggest model,’ but the most useful, personalized, and less frustrating. This shift matters more than any product launch.”
Ad experiments continue in the background, but full rollout is delayed so the team can first improve core UX and personalization for 800M+ weekly users.
“I fear this is benchmaxing over and over”
The skeptical voice matters: if OpenAI obsesses over benchmark scores while neglecting real-world usability, the “code red” becomes theater rather than transformation.
The “Code Orange” Prelude
According to the internal memo, OpenAI had already sounded a “code orange” alarm before Monday’s “code red” declaration. This suggests the company recognized ChatGPT’s problems weeks or months ago but waited until Google’s Gemini 3 momentum forced decisive action.
When GPT-5 dropped in August 2025, user complaints were widespread: it came across too robotic, struggled with basic math, and felt less useful than GPT-4. OpenAI rolled out an update in November to fix tone and responsiveness—but clearly, those fixes weren’t enough.
The Three-Year Reversal
Three years ago, the situation was reversed. Google declared its own “Code Red” in December 2022 to counter the sudden threat ChatGPT posed to Google Search. That internal shake-up led to significant restructuring and eventually to the launch of the Gemini family.
Now, OpenAI faces the same pressure Google did — an insurgent competitor with massive distribution (Google’s integrated search) and rapidly improving technology.
The Star Wars Framing
As one observer noted: “It’s almost like Star Wars; with Google, the empire has struck back. Now it all comes down to this: will we see Return of the Jedi, and can OpenAI beat Google again?”
What Happens Next
In the coming days and weeks:
- New reasoning model launches (next week) with internal claims of beating Gemini 3
- Daily ChatGPT improvement calls across engineering and research teams
- Imagegen upgrades to compete with Nano Banana Pro
- Ads delayed but not canceled—experiments continue in background
- Temporary economic headwinds acknowledged as Google gains market share
The next few months will determine whether OpenAI can defend its crown or whether Google’s structural advantages (search integration, cloud scale, TPU chips) make Gemini the inevitable leader.
“We’re entering a new phase: it’s no longer about ‘the biggest model,’ but the most useful, personalized, and less frustrating. This shift matters more than any product launch.”
But the skeptical voice matters too: “I fear this is benchmaxing over and over, and neglecting real world improvements.”
The truth likely lies in between. OpenAI needs both benchmark wins (to reassure investors and attract talent) and real-world usability gains (to retain users who might otherwise switch to Gemini).
For now, the empire has indeed struck back. Whether we get a “Return of the Jedi” or a “The Empire Wins” ending depends on whether OpenAI’s next reasoning model delivers on Altman’s internal claims—and whether that’s enough to stem Google’s momentum.
The AI race just became a two-horse sprint. And both horses are running flat-out.
