What is GPT-6?
- GPT‑6 is OpenAI’s next major numbered model. It is currently in development and is expected to launch in 2026.
- According to Sam Altman, GPT‑6 will focus on memory, agentic features, and improved multimodal reasoning.
- GPT-6 will be trained on massive compute clusters powered by Stargate and will massively outperform GPT‑5 and GPT‑5.1.
What is a Likely GPT-6 Release Date?
The most likely GPT-6 release date is the first half of 2026.
Despite early speculation about a late 2025 release, OpenAI has confirmed that GPT-6 will not launch in 2025, and that GPT-6 will not be released until sometime in 2026 — that’s all we know officially.
Along those lines, Sam Altman has said that the upcoming model will arrive sooner than the long gap between GPT-4 and GPT-5, and we know that the next generation of ChatGPT is already in development.
Early 2026 is the most likely release date, as it would create a much shorter gap than the 28 months between GPT-4 and GPT-5. This aligns with Altman's comments about transitioning to more iterative release cycles.
This is based on two separate factors:
- Altman has said the team rarely sets targets more than six months ahead.
- The large‑scale hardware required for GPT‑6 won’t fully arrive until the second half of 2026, suggesting the model’s first version may launch before the full infrastructure comes online.
Interestingly, OpenAI is likely already conducting the initial training phases of ChatGPT at its Abilene Stargate campus. Experts believe the model likely exists in some form internally.
GPT-6 Features
Not much is known about the features yet, but Sam Altman said that the future of ChatGPT is likely about long-term memory and customization. Based on that, we can speculate that the company is moving in that direction.
Along those lines, here are the most likely upcoming GPT-6 features and how they’ll change the way you interact with ChatGPT.
GPT-6 Memory
Altman has emphasized that memory is the most important part of the next-generation system. GPT-6 is expected to support:
- Long-term recall of preferences, ongoing projects, and past conversations
- User‑controlled storage, allowing people to manage or delete remembered information
In practice, this means that, rather than every chat feeling like a fresh start, the AI will maintain continuity across weeks or months of conversations and be able to reference previous discussions.
To support this, OpenAI will implement stronger safeguards and better encryption to ensure that conversations remain private.
Do you want your AI assistant to remember more about your past interactions?
GPT-6 Agent Mode
OpenAI will likely continue to develop the chatbot's agentic capabilities. We've already seen improvements in this area with GPT-5.1.
GPT-6 will likely build on this improvement and learn to break goals down into actionable plans more effectively, integrate more external tools and APIs, and become more autonomous. Admittedly, agentic mode is a feature that is more valuable to power users and enterprises. However, it is an area in which OpenAI sees a lot of potential, especially given the intense competition from Claude and Kimi K2.
GPT-6 Personalization
OpenAI will give you more control over how GPT-6 responds to your questions and even its worldview. This means that if you disagree with ChatGPT, including on topics like politics, you’ll be able to customize it. As ChatGPT's user base grows, OpenAI believes that a one-size-fits-all approach simply doesn't work at that scale anymore.
GPT-6 Might be More Multimodal
In addition to text, images, audio, and code, GPT-6 will likely offer improved video reasoning and continuous processing. Currently, you can take a picture and ask ChatGPT questions about it. Imagine being able to record a video and have the chatbot comment on what’s happening in real time while following instructions.
For example, you could write, "Help me fix the bike," film the repair process, and have the chatbot guide you. It sounds amazing, but unfortunately, no details have been confirmed, so we're just speculating.
OpenAI hasn’t revealed exact numbers, but there’s enough reliable reporting to outline what the next model will likely look like.
GPT-6 Technical Specifications
OpenAI hasn’t revealed exact numbers, but there’s enough reliable reporting to outline what the next model will likely look like.
First, researchers estimate that GPT-6 could have multiple trillions of parameters. There are also estimates that OpenAI’s hardware can train a model with 20 to 50 trillion parameters. This would make GPT-6 the densest AI model in the world and significantly larger than GPT-5.
Why this matters: while this is somewhat of an oversimplification, a larger number of parameters essentially means a smarter model, essentially.
Secondly, rumors suggest that GPT-6 will consist of multiple models rather than one giant network. This would include:
- Mixture‑of‑experts (MoE) layers
- More advanced routing and retrieval
Why this matters: This architecture would make the model more efficient. Running a single 50 trillion parameter model would be slow and extremely expensive in terms of computing power.
How is GPT-6 Being Trained?
To train a model that huge, will require over 100,000 high‑end GPUs, including H100 and GB200. There are reports that Microsoft and OpenAI are using so much power to train GPT-6 that they’re pushing the physical limits of power grids, networking, and cooling to make this all possible.
Even deploying more than 100K GPUs in a single region risks straining local power — an indicator of how massive GPT‑6’s training run will be.
That being said, most of GPT‑6’s compute will come from OpenAI's multi‑site Stargate project. With up to 10 gigawatts of planned capacity and billions of dollars in investment, the infrastructure is designed specifically to train frontier models at GPT‑6 scale.
To understand the scope of GPT-6 training, you need to know about two projects: Stargate and Abilene.
What is the Stargate Project?
Stargate is OpenAI’s infrastructure program backed by OpenAI, SoftBank, Oracle, and Abu Dhabi’s MGX. It’s designed to supply the power and hardware needed for frontier‑scale models.
It features:
- Nearly 10 GW of total compute capacity
- More than 2 million chips
To put that into perspective, this is one of the largest private infrastructure projects ever attempted in the United States.
What is the Abilene Data Center?
Abilene is OpenAI’s flagship campus in Texas, where much of the early GPT-6 training work is believed to be taking place. The site already hosts:
- Over 100,000 NVIDIA chips
- Roughly 200 MW of active power across the first two buildings
Even more construction is underway to increase the total power draw by 1.2 GW.
Fun fact: OpenAI began receiving GB200 racks in mid-2025. This means that the earliest phases of GPT-6 development likely began shortly thereafter.
Why This Matters
The main reason large models mostly hit limits is because there isn't enough compute. OpenAI executives have said many times that GPU shortages are the main problem. They already have better models internally, but they deploy them because they don't have enough GPUs. Stargate tries to solve that. They have so much power that they can train GPT-6 faster and in sizes that were impossible before.
Where to Access GPT-6?
Once GPT-6 is available, you will be able to access it on Overchat AI. It will also be available on OpenAI’s official tools, within ChatGPT, and via the OpenAI API. Here’s everything you need to know:
Overchat AI
Overchat already integrates GPT‑5.1, so as soon as GPT-6 releases we will also add it to the platform. In short: you won’t be able to try GPT‑6 until its official launch, but when it arrives, you’ll find it on all major AI platforms, including Overchat AI.
ChatGPT
GPT-6 will most likely be released within ChatGPT. The rollout will likely start with:
- ChatGPT Plus subscribers
- ChatGPT Team and Enterprise users
Historically, OpenAI first introduces new models to its paid tiers before expanding to free users. GPT-6 will likely become available in Overchat AI before it is available to free ChatGPT users.
OpenAI API
Shortly after the ChatGPT launch, GPT-6 will be added to the OpenAI API. This will allow developers to integrate it into apps, agents, and workflows. This is where most people start experimenting.
Bottom Line
OpenAI has recently released the GPT-5.1 model, which is already available in Overchat AI. While this release is an incremental improvement, GPT‑6 is OpenAI’s next major numbered model. It is currently in development and is expected to launch in 2026. OpenAI has confirmed that GPT-6 will not launch in 2025, and that GPT-6 will not be released until sometime in 2026 — that’s all we know officially. Along those lines, Sam Altman has said that the upcoming model will arrive sooner than the long gap between GPT-4 and GPT-5, which places the most likely release date in early 2026.
