Summary

Scaling reinforcement learning (May/2025)

Download source (PDF)
Source: OpenAI (9/May/2025). Permissions: Yes, you can use these visualizations anywhere, cite them.

2026 frontier AI models + highlights

Older bubbles viz

2025Q1 Download source (PDF)
Permissions: Yes, you can use these visualizations anywhere, please leave the citation intact. Download source (PDF)
Permissions: Yes, you can use these visualizations anywhere, please leave the citation intact.

Oct/2024

Download source (PDF)
Permissions: Yes, you can use these visualizations anywhere, please leave the citation intact.

Feb/2024

Nov/2023

Download source (PDF)
Permissions: Yes, you can use these visualizations anywhere, please leave the citation intact.

Mar/2023

Download source (PDF)

Apr/2022

Download source (PDF)

GPT-6 Updates

19/Aug/2025: OpenAI CEO for CNBC:

…he made clear that GPT-6 will be different and that it will arrive faster than the gap between GPT-4 and GPT-5. It won’t just respond to users but will adapt to them, and allow people to create chatbots that mirror personal tastes.

He said he sees memory as the key for making ChatGPT truly personal. It needs to remember who you are — your preferences, routines and quirks — and adapt accordingly.

“People want memory,” Altman said. “People want product features that require us to be able to understand them.”

He said OpenAI has been working closely with psychologists to help shape the product, measuring how people feel while tracking well-being over time. The company hasn’t made that data public, but Altman indicated it might.

He’s looking to the future and brain-computer interfaces. Altman said he finds “neural interfaces a cool idea” and imagines being able to “think something and have ChatGPT respond.”

“There are a few areas adjacent to AI that I think are worth us doing something, and this is one of them,” he said, adding that he’s also interested in energy, novel substrates, robots and faster ways to build data centers.

For now, OpenAI’s core consumer product remains ChatGPT, and Altman said he’s focused on making it more flexible and more useful in daily life. He said he already relies on it for everything from work to parenting questions.

He said, however, that there are limits.

“The models have already saturated the chat use case,” Altman said. “They’re not going to get much better. … And maybe they’re going to get worse.”
(19/Aug/2025)

Jul/2025: Stargate advances:

“Together with our Stargate I site in Abilene, Texas, this additional partnership with Oracle will bring us to over 5 gigawatts of Stargate AI data center capacity under development, which will run over 2 million chips… Oracle began delivering the first Nvidia GB200 racks last month [Jun/2025] and we recently began running early training and inference workloads, using this capacity to push the limits of OpenAI’s next-generation frontier research.” (OpenAI, 22/Jul/2025)

Feb/2025: OpenAI Stargate potential datacentre locations.

2/Feb/2025: OpenAI CEO: 'GPT-3 and GPT-4 are pre-training paradigms. GPT-5 and GPT-6, which will be developed in the future, will utilize reinforcement learning and will be like discovering new science, such as new algorithms, physics, and biology.' (translated from Japanese, 2/Feb/2025, Tokyo)

8/Oct/2024: First DGX B200s delivered to OpenAI for GPT-6 (2025) 'Look what showed up at our doorstep. Thank you to NVIDIA for delivering one of the first engineering builds of the DGX B200 to our office.' (Twitter)

26/Mar/2024: Former Google engineer Kyle Corbitt (Twitter, 26/Mar/2024): 'Spoke to a Microsoft engineer on the GPT-6 training cluster project. He kvetched about the pain they're having provisioning infiniband-class links between GPUs in different regions. Me: "why not just colocate the cluster in one region?" Him: "Oh yeah we tried that first. We can't put more than 100K H100s in a single state without bringing down the power grid." 🤯'

Spoke to a Microsoft engineer on the GPT-6 training cluster project. He kvetched about the pain they're having provisioning infiniband-class links between GPUs in different regions.

Me: "why not just colocate the cluster in one region?"
Him: "Oh yeah we tried that first. We…

— Kyle Corbitt (@corbtt) March 25, 2024

NVIDIA H100 @ 700W x 100,000 = 70,000,000W = 70,000kW = 70MW. Claude 3 Opus said:

70 megawatts (MW) is a significant amount of power. To put it in perspective, this is roughly the amount of electricity needed to power a small city or a large town.

For example, a city with around 70,000 households, assuming an average household consumes about 1 kW consistently, would require approximately 70 MW of power.

And, as detailed in The Memo edition 2/Dec/2023, Microsoft does indeed have an extra 150,000 H100s as of 2023:

18/Mar/2024: CNBC: NVIDIA said Amazon Web Services would build a server cluster with 20,000 GB200 chips. NVIDIA said that the system can deploy a 27-trillion-parameter model… (18/Mar/2024).

Models Table

Dataset

A Comprehensive Analysis of Datasets Likely Used to Train GPT-5

Alan D. Thompson
LifeArchitect.ai
August 2024
27 pages incl title page, references, appendices.

View the report

Timeline to GPT-6

Alan D. Thompson is a world expert in artificial intelligence, advising everyone from Apple to the US Government on integrated AI. Throughout Mensa International’s history, both Isaac Asimov and Alan held leadership roles, each exploring the frontier between human and artificial minds. His landmark analysis of post-2020 AI—from his widely-cited Models Table to his regular intelligence briefing The Memo—has shaped how governments and Fortune 500s approach artificial intelligence. With popular tools like the Declaration on AI Consciousness, and the ASI checklist, Alan continues to illuminate humanity’s AI evolution. Technical highlights.