Code Red: "We're At A Critical Time For ChatGPT." | In The Loop Episode 42

A leaked OpenAI memo dropped just over a week ago, and it's revealed something striking. Sam Altman said, “We’re at a critical time for ChatGPT.” They're playing defense against Google Gemini 3. Prediction markets have shifted dramatically. Google sits at between 87 and 92% odds to have the leading model by the end of this year versus OpenAI at just 8%.

Three years ago, you could never have imagined something like that happening. ChatGPT was just released, and Google announced Code Red. Larry Page and Sergey Brin came back and did all-nighters to review code personally. These were the two founders of Google who had stepped back from operations completely. Teams at Google were reassigned overnight and all resources were reallocated. It felt like the beginning of the end of Google's dominance.

Fast forward to December 2025, and things have changed dramatically. Let's explore the OpenAI situation and see if it's as bad as it seems.

This is In The Loop with Jack Houghton. I hope you enjoy the show. 

‍

‍

‍

What is Code Red at ChatGPT?

If we break down what the leaked Code Red memo says, it's quite interesting. Sam Altman mentioned six big areas of focus for OpenAI: 

1. ChatGPT personalization and customization;
2. Better model behavior measured by the LM Arena benchmarks;
3. Speed and reliability improvements;
4. Reducing over-refusals (which have become a massive area of annoyance for most users on ChatGPT);
5. Launching a new reasoning model (ChatGPT Garlic), they believe, is ahead of Gemini 3;
6. And image generation improvements.

Mark Chen, OpenAI's Chief Research Officer, has become personally accountable to leading all daily progress calls on ChatGPT. This is a refocusing on ChatGPT as a core product. It's massive internal urgency at scale at OpenAI—they've gone from yellow warnings to a red warning, their highest alert level internally.

Yet most coverage online has completely ignored the fact that this wasn't a sudden change to Code Red. Code Red was the escalation of a crisis that began a good few weeks ago.

Back in October, Sam Altman sent an internal memo to various team members warning staff to expect a rough period, maybe even slower financial growth. He admitted in that memo that Google's been doing excellent work recently in every single aspect, particularly in pre-training—an area that OpenAI haven't been able to innovate in for quite some time.

When Code Red dropped in December, it wasn't panic. It was the second alarm bell.

‍

What OpenAI is dropping from their roadmap

What's most telling about this announcement is what they've explicitly said they're going to drop.

All advertising plans are on hold. They were quite far along. Android beta code actually found in late November explicit references to ads features, search ad carousels, and branded content. OpenAI were building serious ad infrastructure, and they forecast they were going to make a billion dollars in free user monetization by 2026, which has been stopped and halted as a project.

To me, this shows how confident they must have been at one point to pursue ads. And that makes me personally sad that ads are going to be brought into any ChatGPT or large language model.

Three years ago, Sam Altman said ads were unsettling and a last resort for a business. Earlier this year, he claimed he loves how Instagram does ads and thought OpenAI could bring some cool product ads into their platform, too. They were clearly putting a massive amount of focus there.

Another area they're dropping is shopping and health agents—AI agents that help you buy healthy things and do healthy things. They've also paused the Pulse feature, that proactive morning briefing system that Sam Altman declared as his favorite feature of ChatGPT.

‍

‍

What the Code Red announcement tells us about OpenAi’s strategy

OpenAI is consolidating its focus and effort. They're abandoning revenue diversification to double down on their core product, which is ChatGPT, and the quality they're providing to users (which has definitely dropped).

Some may say it's a risky bet given the financial pressures they're under and the fact that everybody's talking about AI bubbles and things going wrong. There's a lot of pressure there and they've got incredible financial targets to hit.

However, I'd do the exact same thing at Mindset AI. I'd focus on what is winning most in the market and become the best in the world at that thing because ultimately if that foundation dies off, it's game over.

Steve Jobs famously said that it's better to have fewer products but do them incredibly well and do incredible marketing for those products than having lots of random products in the market that confuses everybody.

‍

The financial pressures

OpenAI is burning through cash at a crazy rate. $5 billion in losses in 2024, projected to be $8.5 billion this year. HSBC analysts forecast a $207 billion funding shortfall—they're short $207 billion from now to 2030. There were apparently some leaked financial figures that they expect to lose up to $74 billion in a year.

The company's also committed to $1.4 trillion in infrastructure spending over the next eight years.

This is tough because data center infrastructure needs to be built out to hit your growth targets. They'll look at the amount of users or products in the market, therefore how much data center capacity do they need. Based on the products they want to launch, they're going to need more data centers and GPUs. That creates a fixed amount of spending on data warehouse infrastructure. If they don't commit to it now, they won't have it in the future because it's not something that you just buy like a car and drive away with. It needs to be built out for you.

These are things they can't escape as a financial and business model, but gosh, it makes it very expensive to run.

They're projected to need $200 billion in revenue by 2030 just to become profitable. Their current run rate is $13 billion.

A run rate is a quick and straightforward financial projection that takes your current short-term revenue and earnings, say monthly or quarterly earnings, and then uses them to guess what your total yearly earnings would be.

The math just doesn't work out. They're not earning enough money per month and per quarter to ever hit that amount. They're going to have to just keep growing bigger and bigger and faster and raise more and more money to do it. The way that's going to work out is having market dominance. If they don't have market dominance and people don't believe in the product, they're not going to give them much money.

Their valuation jumped from about $157 billion in October 2024 to double that—$300 billion by March of this year. Then by October of this year it was at $500 billion.

For what is technically a startup, that is a 38 times multiple on how much they earn. For context, Nvidia, with huge earnings, is trading at 24 times their revenue. OpenAI is valued at 38 times their revenue.

To achieve these crazy numbers and market narrative and the amount of money that people want to put into them, they have to be a dominant player in the market. Gemini 3 coming along and potentially beating them is an existential threat, mainly because all of their competitors have access to better credit than they. Google can get money more easily and quickly than OpenAI.

‍

User feedback and the over-refusal problem

One developer on the OpenAI community forum said that because of all the new changes, ChatGPT as a subscription is utterly useless in most cases because either when they're trying to do coding tasks, it is overly verbose, but people who are trying to do writing tasks it's not verbose enough.

It's also got a big problem with over-refusals, which is when OpenAI explicitly refuses to do things for you. If you ask it what the very best biscuit type is, it will say that you shouldn't be biased towards biscuit types.

One user asked for an explanation of Russian roulette, and ChatGPT refused to provide the rules, even though the question was posed out of curiosity. I tried to do this over the weekend as a test. I wanted a series of questions of what would you do in this situation and I wanted to make them either scary or crass and rude and you just couldn't get it to work.

The number one biggest problem is still sycophancy—the model just agreeing with everything somebody is saying and refusing to push back or giving false validation. OpenAI say they're going to be focused on personalization and customization, so they're trying to fix this, but most people are wanting a model that is going to disagree with them.

‍

The switching problem

Another major problem OpenAI has here is the switching problem.

One Reddit user perfectly captured this: "Claude smokes GPT for Python, and it isn't even close on my end."

Another said that they switched to Claude recently and it enabled them to build an entire phone app in a day. Because of the context window, it required you to press continue four separate times—it ran out of context and then needed to start again yet perfectly started where it left off and did a good job.

The benchmarks back all of this up. On the SWE benchmark, which is the gold standard for evaluating coding, Claude Sonnet scored 72% and GPT-4o scored 33%.

It also shows in how the world has adopted these different language models. Cursor adopted Claude as their main programming language. GitHub did the same. Windsurf, SourceGraph, Cody, Augment Code, they all did the same.

The financial impact of this is pretty massive. Claude Code has hit a billion dollars in revenue in six months. Anthropic is out there acquiring a series of different businesses to make their Claude Code product even better. They're not just winning the market or ahead in the market. They are investing extremely heavily into becoming the leader and winner of that market.

You'll notice that in the list of priorities Sam Altman laid out, coding wasn't included.

My read is that OpenAI is quietly accepting that the market has been lost for now to coding to Anthropic. They've still got their Codex product out there. They're probably going to still continue it, but clearly it's not the main priority anymore. It makes sense because they're having huge pressure from the consumer market when it comes to Gemini. That's their main base of users and revenue—800 million ChatGPT users.

‍

Competition: Google and Anthropic

OpenAI is facing huge competition and I think it's worth spending a little bit of time on that competition.

On benchmarks, Google's Gemini 3 currently sits at number one on the LMArena with a record 1,491 score. Anthropic dominates six of the top 10 spots, and ChatGPT's o1 is only in sixth place. That's not terrible, but it's also not market leadership.

On user metrics, ChatGPT still has 800 million weekly active users. According to OpenAI, they dominate with 70% of all AI assistant usage.

That's not going to last forever.

Gemini just jumped from 450 to 650 million monthly active users. Quite far behind OpenAI, but they will and are able to catch up. That happened between July and October of this year.

A metric that will worry OpenAI is session time. Google now has a better session time than OpenAI—7 minutes and 8 seconds versus 6 minutes and 25 seconds at OpenAI. When users are spending more time with one product over another, it probably will start to tell you something. Unless the extra couple of minutes in Gemini is spent being frustrated, which I don't think is the case.

ChatGPT's daily active users dropped 6% recently.

They're also facing massive competition from an enterprise market perspective. According to a mid-2025 report, Anthropic now has 32% of enterprise market share versus OpenAI's 25%. Just one year prior, OpenAI had 50% and Anthropic just had 12%. That's a massive reversal.

‍

The talent drain

OpenAI is facing a massive talent drain. Dozens of top researchers have left for competitors. OpenAI's CTO launched Thinking Machines and recruited loads of incredible talent from OpenAI. Meta's new superintelligence lab poached people aggressively too.

Mark Chen reveals how intense this talent war is, noting that Meta spends $10 billion annually on acquiring AI talent. Apparently, Mark Zuckerberg has been hand-delivering people soup to try to persuade them to come on board.

No wonder that OpenAI has sounded the alarm and made Code Red such a big priority.

‍

‍

Four things to watch

Over the next period, I think you need to be watching for a few things.

‍

1. ChatGPT Garlic

The Code Red memo promised a new reasoning model, called ChatGPT Garlic, over the next few weeks that would be better than Gemini's latest model. Mark Chen, the Chief Research Officer, said that OpenAI had succeeded in infusing a smaller model with the same amount of knowledge that had previously been required for a larger system. 

This is important because when they throttle intelligence—reduce the number of tokens and make everything concise—it just makes you feel like it's stupider. Every single person wants the most powerful, cleverest model possible all the time.

‍

2. User retention

ChatGPT's engagement metrics are still incredibly strong—12.74 visits per unique visitor, but Gemini's user base is growing quickly. From 450 million to 650 million in just a few months. People will be watching very closely to see if OpenAI's improvements in their product leads to a more retained user base or just a slower attrition and a slower decline.

‍

3. Enterprise market

A third big thing to watch for will be enterprise wins. The Fortune 500 market is significant to these products, and if you see Anthropic winning with enterprise organizations, it will tell you a lot about where OpenAI is heading.

‍

4. Developer sentiment

The coding market is where AI accomplishes big long-term value through API revenue, tool integration, developers building products on their platform. Claude is owning that conversation. OpenAI has to close the coding gap.

‍

‍

‍

‍

Closing thoughts

What I find interesting about this moment in time is that just three years ago, smart money would say that Google was going to struggle. They'd been caught completely flat-footed. The founders had to come back. The panic was real. Market sentiment was very negative.

Yet today, Google has the best-ranked model in the world, the fastest growing user base in the world, and the infrastructure advantage from building their own chips, their own products—just a vertically integrated product set.

OpenAI are now in a position that Google were in just a few years ago. The difference is OpenAI doesn't have Google's resources. They don't have distribution. They don't have the hardware vertical integration and they're burning through capital at a crazy rate. They're having to fight a consumer war against Google yet they're fighting a coding war against Anthropic and an enterprise market war against Microsoft. These are all very different markets with very different dynamics.

OpenAI's Code Red is a reminder that perceived market leadership never lasts forever. However, Google's return also shows that anything is possible.