OpenAI's Spud: The GPT-6 Era Is Coming!

OpenAI just finished building their next big model, it's called Spud, and Sam Altman, the CEO of OpenAI, stood in front of his entire company and told them this model is going to really accelerate the economy, not in a press release, not in a tweet, in a room full of his own people. That means something, let me walk you through exactly what we know, why OpenAI made some brutal decisions to get here, and what Spud might actually be when it drops. On March 24th, 2026, journalists Stephanie Palazzolo and Amir Efrati at The Information broke the story. OpenAI has completed initial development of a new model, internally codenamed Spud.

Altman wrote to staff that things are moving faster than any of us expected. He said the release is coming in a few weeks, mid to late April 2026, that's the window. Now here's where it gets interesting, and this is the part that tells you how seriously OpenAI is taking this. To build Spud, to free up the GPU computing power needed, OpenAI actually ended Sora, their AI video generator, completely shut it down, shut down the app, shut down the API, removed the video features entirely inside Chachibot, they also cancelled a major partnership with Disney, a deal involving over 200 characters across Marvel, Pixar and Star Wars, gone.

CFO Sarah Fryer went on CNBC and said it directly, we are just facing a lack of computes, we're having to make really difficult decisions. They walked away from one of the most respectable and recognisable IP libraries in the world, because Spud needed the computing resources more. That's not a routine product launch decision, that is a company clearing the deck, removing everything that isn't essential, focusing everything on one thing. And the leadership restructure that happened at the same time, tells the same story.

So Altman has stepped back from directly overseeing safety and security. Not because they don't matter, but because he needed to personally focus on what he described as building data centres at unprecedented scale. So Chief Research Officer Mark Chen is going to handle safety And here's the detail I keep thinking about, the entire product organisation was renamed, it used to be called Product, now it's called AGI Deployment, not product, not features, AGI Deployment. That name is not an accident, that is a company that believes it's deploying something significant, the kind of thing you name carefully.

So what is Spud actually going to be? Nobody outside OpenAI knows for certain, no capabilities have been confirmed publicly, no benchmarks, no architecture details, nothing. But there are three threads in the reporting around it that when you pull them together, paint a pretty clear picture of what's coming. Let's go through each one.

The first is the audio connection. In January 2026, the information reported that OpenAI had quietly merged multiple engineer research and product teams into one focused effort, to build a completely new audio model from scratch. Not an upgrade to what exists, a full rebuild. A model that speaks more naturally, handles interruptions without breaking, and can talk whilst you're talking.

No more robotic pauses, no more awkward gaps. The target latency was under 800 milliseconds. Now, sub 800 milliseconds is roughly the speed of a natural human back and forth. And that project was on track to release by the end of Q1 2026.

That window lines up exactly with where Spud is landing right now. The person leading that effort is Kundun Kumar. He co-founded Lyrebird AI, a voice cloning startup back in 2019. He holds a PhD in neural audio generation from MILA, under Yoshua Bengio, who's one of the people who built the foundations of modern deep learning.

OpenAI recruited Kumar from character.ai in the summer of 2025. They also brought in Ben Newhouse to rebuild the infrastructure underneath all of it. And these are serious specialist hires for a project that is clearly central to the strategy, not a side experiment. Audio is not a feature they're adding.

It looks like it's becoming a core part of what Chattopadhyay actually is. The second piece is native modality. Right now, when you use Chattopadhyay with voice, it's switching between different and separate systems in the background. One handles your speech, another processes the text, another generates a reply, another converts it back to audio.

You can feel the handoffs if you pay attention. It works, but there's like a mechanical quality to it. The different parts are stitched together rather than built as one thing. Insiders and less reporting suggests Spud may be trained across text, audio, and images simultaneously from the very beginning.

Not fine tuned for each mode separately after the fact, natively multimodal from the ground up. If that's accurate, it changes the experience in a way that's hard to explain until you feel it. A model built this way doesn't need to translate between modes. You show it something, talk to it, and it responds to the full picture at once, not a sequence of steps running one after the other.

That's a totally different kind of AI to interact with. It feels like a tool, and more like something that actually understands what's happening. The third piece is the super app. So, the Wall Street Journal reported on March 19th that OpenAI is building a single desktop application, merging ChatGBT, their coding tool Codex, and a new browser called Atlas into one product.

One login, one interface. President Greg Brockman is leading that personally, which tells you how important it is internally. The concept is pretty straightforward. You browse the web, write a brief, generate copy, analyze data, have a conversation, and it's all powered inside the same product.

Powered by one model that understands all of it, simultaneously. Spud will be the engine underneath that whole thing. If Spud is natively multimodal with rebuilt audio, the super app becomes much more interesting, because then it's not just combining tools. It's giving you one thing that sees, hears, reads, and acts.

All of it at the same time. Now, is Spud actually GPT-6? Well, it's probably not the final version, but it's probably a step along the way. Here's the reasoning.

Altman separately told staff that OpenAI is training a model at their facility in Abilene, Texas. Part of a huge infrastructure that he believes will be the best model in the world, hopefully by a lot. And that sounds like a totally different, much larger model, that's still in training. Prediction experts are saying that as of late March, they reckon about 2% probability that a model officially called GPT-6 will land next month in April 2026.

So Spud is most likely GPT-5.5, or a new naming scheme entirely. A significant step that comes before the even bigger generation currently being trained. But don't let that framing shrink what Spud might be. A natively multimodal model with a rebuilt audio architecture and a unified super app wrapped around it, is still a huge jump from what exists today.

And the version number is much less important than whether it actually works differently in your hands. Now let me give you the competitive picture. Because understanding why OpenAI is urgently moving is part of understanding why Spud matters so much. So OpenAI actually declared internal code red in December 2025.

Not publicly, internally. Product chief Fiji Saimo told staff directly, we are very much acting as if it's a code red. She described gains by competitors as a wake-up call and told the team to stop chasing side projects. The focus had to come back to core.

And why the urgency? Because the gap between OpenAI and its competitors closed way faster than expected. And in the areas that matter to most serious entrepreneurs and businesses, that gap is now reversed. For example, Anthropx Claude is currently leading the models for complex enterprise work.

On SWE Bench, which measures whether AI can solve real problems and real engineering problems for software, Claude Opus scores 80.8%. That's the 4.6 model. That's the best score in the world right now. Companies that have moved toward Claude consistently report the same reason.

It stays on task across multi-complex tasks. It's great at multi-step work and it makes fewer hallucinations and errors along the way. And that's really difficult to close in terms of reliability over months of real usage. I mean, Google as well, they have 650 million active users for Gemini each month.

Their Gemini 3 DeepThink model scores 84.6% on ARCAGI2, an abstract reasoning test specifically designed to be hard for AI systems. The highest score of any model anywhere. And Google is a company that's not slowing down. And then you've got DeepSeq as well.

You've got all these Chinese companies that are releasing these models that are open-sourced or much cheaper, easier to run, and they work locally. And this is the competitive environment that Spud is walking into. It's not a market that OpenAI comfortably dominates anymore. It's a market where multiple serious competitors are each winning on something different.

That could be reliability. It could be open source. It could be reasoned depth. Spud needs to give people a reason to use OpenAI again as a default.

Now, here's what this means for you right now. I would be aware that you know, ChatGPT may be hyping up these models, but quite often when they get released, they struggle. And I'll give you an example. There was so much hype leading up to ChatGPT 5, but when it actually came out, it struggled.

You know, it wasn't as hyped up as it expected. Everyone was expecting AGI. And the reality was that in many ways, people actually preferred GPT 4.0 to GPT 5. So Spud may be the next jump.

GPT 6 may be the next jump. But the difference here is that when it comes to these new models, I would keep your expectations low with ChatGPT. I would keep your options open as well. I wouldn't build all your infrastructure or all your AI automations around one API at this point.

You know, pick the right model to what works the best. Spud is the next jump, and the window between now and when it lands is time to build that fluency. Inside the AR Profit Board, we show you how to use tools like this, how to grow your business, how to save time with it. And when Spud drops, we'll run dedicated sessions on what's actually changed, how to use the new capabilities, and how to build it into your workflow fast.

You also get daily tutorials, 30-day roadmaps. And the whole goal of this is to get you more leads, more customers, and help you grow and save time with AR automation. So that's inside the AR Profit Boarding. Link in the comments description, or just go to the AIProfitBoarding.com.

So when Spud drops, what should you actually pay attention to? Because launch day is going to be loud. Every lab publishes all these crazy benchmarks, promises the world, and then when you test it, not so good, right? So I would actually test the audio version of the next model that comes out.

That would be very interesting. I would also keep your options open, and I wouldn't focus all of your automations on just one audio. I would also focus on performance rather than benchmarks. When these new tools come out, they're exciting, and there's a lot of noise, but when you test them out, they're not as good as you think they are.

So whether Spud delivers on what we expect, we'll find out in a few weeks. But what's true either way, is that the AI tools available right now, before Spud, before the next jump, are already powerful enough to help you build awesome generations, awesome automations, right? And there's a window now between this month and next month when the next version of ChatGPT comes out. And you don't need to wait for any new models to come out to really get value out of these AI automations, right?

So for example, for us, we can easily automate content, we can easily automate SEO, and we don't need to wait for the next model. But when it comes out, will I be testing it? Absolutely. Am I looking forward to it?

Absolutely. So either way, hope to see you inside the AI Profit Building. Thanks very much for watching, and I will see you in the next one. Cheers, bye-bye.