OpenAI Just Told Us All About GPT-5
17:21 Share

Today on the AI Daily Brief, we finally learn what GPT-5 is going to be. Before that, in the headlines, Anthropic projects some serious revenue gains. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes. ♪

Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around five minutes. Well, it turns out that Anthropic is expecting some extremely rapid growth over the next few years. The startup, as we have previously discussed, is currently courting investors and as part of that are sharing some fairly eye-popping forecasts.

According to leaked documents obtained by The Information, the company's base case is reaching $12 billion in revenue by 2027. That would be a big jump from last year's numbers, which were estimated at around $600 million. They also gave an optimistic projection, showing the company hitting $34.5 billion in revenue by 2027. That would represent a huge catch-up to OpenAI, which generated five times the revenue of Anthropic last year, but only, and those are in a big air quotes only, forecast $44 billion for 2027.

Now, more interesting, I think, than just the projections on their own are where they believe the growth is going to come from. The company has experienced significant growth over the last year. They had around $100 million in annual revenue for 2023 and saw monthly earnings grow from $8 million last January to $80 million by December. A big part of driving that was AI coding tools hitting mass adoption, specifically the explosion of usage of Cursor, through which many, if not most people, use Cloud 3.5 Sonnet.

For whatever reason, that model seems to be just the preferred model when it comes to coding assistants. And that's obviously a huge, huge use case for AI right now.

When it comes to the breakdown of where this revenue is going to come from, Anthropic says that they expect API revenue to hit $20 billion by 2027, which is three times what OpenAI forecast for that year from their API revenue. Even in their base case projection, they have API revenue beating OpenAI's API revenue. In terms of the specifics of the raise, it appears that Anthropic is trying to raise $2 billion at a pre-money valuation of $58 billion.

Still, for the moment, these companies are juicing through money. Anthropic is projecting to burn $3 billion this year, which would be significantly less than last year, which saw a burn of $5.6 billion. Interesting stuff in how these companies see the future shaking out. One other Anthropic-related story, that company's CEO, Dario Amodai, has urged AI researchers to pick up the pace when it comes to understanding AI.

Earlier this week, he called the Paris AI Summit a missed opportunity and urged, quote, greater focus and urgency needed on several topics, given the pace at which the technology is progressing. One of the things that was interesting about his comments is that he tried to shift away from the bifurcation of painting AI policy as a fight between safety on the one hand and opportunity on the other, as Vice President J.D. Vance had, of course, framed it, and instead effectively was arguing for the acceleration of research and thinking on the governance side as well.

i.e. instead of deceleration on the pace of technology, he wants an acceleration on all the things that go around the technology. At a developer-focused side event, he said, It's definitely a race. It's a race between making the models more powerful, which is incredibly fast for us and incredibly fast for others, and our understanding has to keep up with our ability to build things.

Moving over to big tech for a moment, even the Apple intelligence features that were so far somewhat disappointing have not been available to iPhone users in China. That's because that country requires Apple to partner with a local builder on their iPhone models, and they've been in the process of selecting that partner. They had apparently initially selected Baidu, but that company struggled to adapt their larger models for iPhone use. Subsequently, they reviewed a number of companies, including apparently DeepSeek and ByteDance, but ultimately decided to go with Alibaba.

They are, of course, hoping that this will drive an increase in sales. During last month's earnings call, CEO Tim Cook blamed a lack of Apple intelligence for an 11% drop in Chinese sales. One more interesting one on Apple. According to longtime Apple insider Ming-Chi Kuo, Apple is, quote, exploring both humanoid and non-humanoid robots for its future smart home ecosystem. He said the products are likely still in the proof-of-concept stage and therefore wouldn't hit mass production until 2028 at the earliest.

While many robotics companies are debating the merits of humanoid versus non-humanoid designs, Apple appears to be thinking differently. You might have seen a demo video shared by Apple last week which showed a robotic lamp. Taking a page out of the Pixar style guide, the lamp had human-like movements and gestures. The demo showed the lamp following as a person moved a book around a table, while a user could also wave to grab the lamp's attention and gesture for it to point at the wall.

In an accompanying research paper, Apple described this as non-anthropomorphic design. The logic is that a robot who responds to gestures like a human could present a more natural UX than voice commands.

Finally today, we always keep track of how companies are trying to bring chips in-house. And on that front, Meta is in talks to acquire South Korean chip startup Furiosa AI. According to Forbes reporting, the acquisition could be announced as soon as this month. Furiosa was founded in 2017 by former Samsung and AMD employees. They develop inference chips that could speed up the performance of Meta's models. Furiosa claims their RNGD chip offers three times better performance per watt than NVIDIA's H100.

There you have it. That is going to do it for today's headlines. Next up, the main episode. Today's episode is brought to you by Vanta. Trust isn't just earned, it's demanded. Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in.

Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC 2 and ISO 27001. Centralized security workflows complete questionnaires up to 5x faster and proactively manage vendor risk. Vanta can help you start or scale up your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building your company.

Join over 9,000 global companies like Atlassian, Quora, and Factory who use Vantage to manage risk and improve security in real time.

For a limited time, this audience gets $1,000 off Vanta at vanta.com slash nlw. That's v-a-n-t-a dot com slash nlw for $1,000 off. If there is one thing that's clear about AI in 2025, it's that the agents are coming. Vertical agents by industry, horizontal agent platforms.

Agents per function. If you are running a large enterprise, you will be experimenting with agents next year. And given how new this is, all of us are going to be back in pilot mode.

That's why Superintelligent is offering a new product for the beginning of this year. It's an agent readiness and opportunity audit. Over the course of a couple quick weeks, we dig in with your team to understand what type of agents make sense for you to test, what type of infrastructure support you need to be ready, and to ultimately come away with a set of actionable recommendations that get you prepared to figure out how agents can transform your business.

If you are interested in the agent readiness and opportunity audit, reach out directly to me, nlw at bsuper.ai. Put the word agent in the subject line so I know what you're talking about. And let's have you be a leader in the most dynamic part of the AI market. Hello, AI Daily Brief listeners. Taking a quick break to share some very interesting findings from KPMG's latest AI Quarterly Pulse Survey.

Did you know that 67% of business leaders expect AI to fundamentally transform their businesses within the next two years? And yet, it's not all smooth sailing. The biggest challenges that they face include things like data quality, risk management, and employee adoption. KPMG is at the forefront of helping organizations navigate these hurdles. They're not just talking about AI, they're leading the charge with practical solutions and real-world applications.

For instance, over half of the organizations surveyed are exploring AI agents to handle tasks like administrative duties and call center operations. So if you're looking to stay ahead in the AI game, keep an eye on KPMG. They're not just a part of the conversation, they're helping shape it. Learn more about how KPMG is driving AI innovation at kpmg.com slash US.

All right, friends. For basically the entirety of 2024, one of the big questions was when are we getting GPT-5 or at least GPT-4.5?

When ChatGPT first launched in November of 22, it was only a few months before we got GPT-4, and then for the entirety of 2023, GPT-4 was solo out in front of everyone else, and most people kind of anticipated that as the rest of the world caught up with GPT-4 at the beginning of 2024, that OpenAI would once again zoom ahead with GPT-5 or at least GPT-4.5.

That of course never happened, and instead what we got towards the end of the year was an actually different approach to models, which OpenAI called reasoning models. We got the earliest versions of O1, and OpenAI started to indicate in places like Reddit AMAs that they were really going to be doubling down on reasoning models as the go-forward approach.

Now, throughout all of this, we've gotten a million different names, GPT-4.5, GPT-5, Orion, O1. And it appears as though Altman and OpenAI have decided that enough is enough in terms of the complexity and confusion. And they want to just actually be transparent about where things are headed.

On Wednesday of this week, Altman took to X to write, We want to do a better job of sharing our intended roadmap, and a much better job simplifying our product offerings. We want AI to just work for you. We realize how complicated our model and product offerings have gotten. We hate the model picker as much as you do and want to return to magic unified intelligence. As a quick aside, the model picker is this thing at the top, which now has GPT-4-0, GPT-4-0 with scheduled tasks, 01, 03-mini, 03-mini-high, 01-pro mode, GPT-4-0-mini, and GPT-4.

Coming back to Sam, he writes, we will next ship GPT-4.5, the model we call Orion internally, as our last non-chain-of-thought model. In other words, as our last non-reasoning model. After that, a top goal for us is to unify O-series models and GPT-series models by creating systems that can use all our tools, know when to think for a long time or not, and generally be useful for a very wide range of tasks.

In both ChatGPT and our API, we will release GPT-5 as a system that integrates a lot of our technology, including O3. We will no longer ship O3 as a standalone model. The free tier of ChatGPT will get unlimited chat access to GPT-5 at the standard intelligence setting, plus subscribers will be able to run GPT-5 at a higher level of intelligence, and pro subscribers will be able to run GPT-5 at an even higher level of intelligence. These models will incorporate voice, canvas search, deep research, and more.

All right, so that is the specific update. One of the first responses was perhaps a tongue-in-cheek joke about how this didn't necessarily super clear things up. Bloomberg's Joe Weisenthal wrote, OpenAI knows its model offerings are confusing to people. Going forward, OpenAI is simplifying it by replacing GPT-4 with GPT-4.5 Orion, final non-chain of thought, and then GPT-5, including O3 at standard plus pro-intelligence tiers featuring voice canvas search and deep research. Still, there are a couple of really big things here.

First of all, the fact that we're finally getting GPT 4.5 in what it appears like a matter of weeks is something that people are really looking forward to. Yes, the reasoning models have opened up new capabilities, but there are a lot of things that the non-reasoning models are still better for, but that people still want to see performance gains in.

Also, when it comes to this idea of the model picker being a bad user experience and wanting to simplify that, I think this is sort of self-evidently true in that even for someone who's super up on this, having to pick between these different models for every different task is not necessarily a great experience, although it does give you a lot of pinpoint flexibility and the ability to test different models against different prompts. But the move to simplification also reflects, I think, where ChatGPT sits in terms of its mainstream appeal.

ChatGPT is now something like the sixth most trafficked website in the US. It is a major consumer product, not just an enterprise product. It has that Kleenex sort of brand where for many people, AI is literally just ChatGPT. That's what they refer to AI as. This push for simplification suggests that that is an audience that they really want to make sure has the ability to wrap their head around and get the most out of these core tools.

This, I think, is validated by the fact that all levels of subscribers are going to have some access to the most advanced models, not just the people who are paying. You gotta think that the Just Work idea also had something to do with the release of DeepSeek, whose app, including the transparency of how it was coming to the conclusions that it was coming to, became super popular, even displacing ChatGPT from the App Store rankings. Remember, with the DeepSeek app, users were presented with a single powerful reasoning model that wildly outperformed what they were used to in their experience with the free tier of ChatGPT.

Moving to the technical side of things, though, I think strategically there's something really interesting about this, even though I don't think that the move comes out of left field. When O1 was first introduced, it was very clearly presented as a separate branch of AI models. This was not GPT-4.5. It was something that was different. The intention, at least at that time, appeared to be to continue to develop along both branches simultaneously.

Now, for those who are watching closely, it wasn't long before that seemed to be moving in a different direction, with, like I said, Reddit AMAs seeming to suggest that the company was much more focused on the reasoning models than they were on the previous approaches. And of course, lurking behind all of this is this question of whether pre-training in general as a scaling strategy has hit a wall.

Google's Gemini 2.0 Pro is the only flagship LLM to be released without reasoning in the past few months, and it doesn't show a step change in performance. GPT-4.5 or XAI's Grok 3 might prove the thesis wrong, but it may be that these labs are just deciding that they're coming up against the ends of their ability to scale pre-training. That leaves these reasoning models or test-time compute models as the new big scaling vector for model improvement moving forward.

And again, for those watching closely, they've already dabbled with the idea of giving different tiers of subscribers different level of test time compute and hence different levels of model performance. Specifically, O3 Mini has low, medium, and high intelligence settings corresponding to longer inference time.

One of the questions for many people is the integration of the experience. In discussing the idea that GPT-5 is going to be a unified model, friend of the show Swix asked OpenAI's Kevin Wheel, in GPT-5, are GPT and O still separate models under the hood and you're making a model router? Or are they going to be unified in some more substantive way? Kevin responded, unified.

This is exactly what Swix had been hoping to see. He said, I felt that the worst timeline was the one that others appear to be going towards, which is train different models and have a model router to create a semblance of AGI-ness. GPT-5 seems to be doing O5 plus GPT-5 merge right, not just simple router of reasoner versus non-reasoner. This is exciting architecture work if more details are to be published.

There are also some really interesting technical challenges here. Automatically selecting the best inference time for a given query, rather than leaving that choice to the user, is something that's going to take some design. Meta recently produced some academic work on the subject, although even that seems quite preliminary.

Now, there also seems here to be a shift in the philosophy of the availability of models for OpenAI. Once again, it feels like reflecting pressure from, among others, deep research. Up until very recently, the latest models had been mostly gated behind subscriptions. Indeed, for some of their real high-end features like deep research, these things were even gated to the $200 per month pro tier. Respectively,

Responding to someone saying that deep research is worth $1,000 a month to them, Altman commented yesterday, I think we are going to initially offer 10 uses per month for ChatGPT Plus and two per month for the free tier with the intent to scale these up over time. It probably is worth $1,000 a month to some users, but I'm excited to see what everyone does with it. Basically, it seems to suggest that going forward, the company is going to try to endeavor where at least even a little bit financially reasonable to give everyone access to the most advanced intelligence, even if not all that much of it.

Some people love the mood. Google AI Studios product lead Logan Kilpatrick wrote,

hence doing 2.0 flash thinking. However, Garrett of DeepRider AI wrote,

Presumably getting GPT-5's integration right means actually being able to speak to both of these use cases, but it does reflect the concern of prioritization of the reasoning models over everything else. And yet that clearly feels like what's happening. Professor Ethan Malek writes, Seems like increasing evidence that LLMs that are not reasoners are going to fade away. There's the explicit announcement from OpenAI, the academic studies on the topic, the fact that Gemini 2.0 Pro and apparently Grok 3 seem to be outclassed by their reasoning AI cousins, etc.

To anyone who isn't following closely, this doesn't matter. They will just know AI gets better. So that is the update we got. Interesting and exciting stuff. Like I said, I am among those people who is super excited to get the actual GPT-4.5 and see how it improves the capabilities around a lot of the day-in, day-out use cases that we have that don't require that reasoning approach. And of course, as soon as it is available, we will get in there, test it, and share what we find with you.

For now, though, that is going to do it for today's AI Daily Brief. Appreciate you listening or watching as always. And until next time, peace.