Image above generated by AI for this analysis (GPT-4o image gen)¹Image generated in a few seconds, on 5 May 2025, via GPT-4o image gen, text prompt by Alan D. Thompson.

Alan D. Thompson
May 2025 (placeholder)

Summary

Organization	OpenAI
Model name	o5 (OpenAI model number five)
Internal/project name	–
Model type	Multimodal
Parameter count	See below: Size estimate
Dataset size (tokens)	See report: What’s in GPT-5?
Training data end date	May/2025 (est)
Training start date	Jun/2025 (est)
Training end/convergence date	Aug/2025 (est)
Training time (total)	– See working, with sources.
Announce date
Release date (public)
Paper	–
Playground	–

Updates

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)

16/Apr/2025: o3 and o4-mini released.

23/Jan/2025: Brad Lightcap, OpenAI (Cisco AI Summit, 23/Jan/2025)

The great insight that Ilya had way back when is that the systems that didn’t seem to work at small scale weren’t fundamentally flawed, they just needed to be scaled up. And we’ve seen demonstrably now that when you scale GPT models, you get consistently better results. We think that continues. Now you have this entirely new vector of scaling, which is the o series of models. We’re very excited about that, and we don’t see any asymptote on scaling laws…

With scaling the big pre-trained transformers, you enter into a regime where you become data limited if you’re not already compute limited. And so in some sense, we hit a wall. It wasn’t a technical wall, it was just an environmental wall. So we could continue to scale those models. There’s no scientific reason why a GPT-5 or 6 couldn’t exist and have the same kind of return to scale that you see going from GPT-3 to GPT-4…

The beautiful thing about RL [reinforcement learning], though, and the way that we’ve built the reasoning models, is they take you on a different path. Someone on our research team described it to me the other day. o1 is almost like a portal to GPT-7 or GPT-8. It’s a way to leverage the knowledge in the pre-trained model, which is a GPT-4 scale model, in a way that basically gives you the effective compute of what a GPT-7 or GPT-8 would otherwise use, to achieve that same level of intelligence. It’s a pure discontinuity on the scaling…

I think we’ll continue to do both [GPT and o models]. I think we still see a lot of potential in pre-training levels, where you’ll see a 5 and a 6 and a 7 coming on that front as well. I think we’ll see where that all goes. Believe it or not, it’s actually almost more continuous than these discrete jumps now.

Major points

Model name

OpenAI (17/Oct/2024 timecode 21m11s, transcribed by Whisper):
We plan to continue developing and releasing models in the new OpenAI o1 series, as well as our GPT series. In addition to model updates, we expect to add web browsing, file and image uploading, and other features to [o1 to] make them more useful in use cases in ChatGPT. And while even today you are able to switch between models in the same conversation, like you saw in the demo, we’re working to enable ChatGPT to automatically choose the right model for your given prompt.

Smarts

The GPT-4x and o Model Family: Varied Intelligence Scores 2024–2025

Scaling reinforcement learning (May/2025)

Reasoning model scores: o1, o3, o4

Size estimate

…

Dataset

The o5 dataset is expected to use much of the initial GPT-3 dataset as detailed in my report What’s in my AI?, and to be very similar to the dataset used for GPT-5, with new synthetic data and partnerships as outlined in my GPT-5 dataset report.

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

Use cases

Coming soon…

2025 frontier AI models + highlights

Older bubbles viz

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)

Models Table

Timeline to o5

Videos

…

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.

• 1
  Image generated in a few seconds, on 5 May 2025, via GPT-4o image gen, text prompt by Alan D. Thompson.
• 2
  Human average, not expert average
• 3
  PhD human average, but not PhD in subject field average
• 4
  "2024 AIME I and II median score=5/15": artofproblemsolving.com, disregards that these humans are probably in the top 1% of mathematics performance: 'Qualification parameters for the AIME depend on the results of the AMC 10 and AMC 12 competitions. For the AMC 10, at least the top 2.5% of all scorers from the A and B competition dates are invited, and for the AMC 12, at least the top 5% of all scorers on each version are invited [to take the AIME test].maa.org
• 5
  https://codeforces.com/blog/entry/126802
• 6
  Human expert average, that is junior developer estimate=10-25%: https://github.com/All-Hands-AI/OpenHands/issues/1693
• 7
  NYU: "Independent samples t-tests suggest that evaluation tasks are significantly harder for people than training tasks... We estimate that the average task accuracy after three attempts on the evaluation set is 64.2% (SD=22.8%, [55.9%, 68.9%]). In addition to this result, we report a first and second attempt average task accuracy of 47.8% (SD=23.2%, [41.6%, 54.6%] and 60.2% (SD=23.3%, [52.4%, 65.4%]) respectively." https://arxiv.org/pdf/2409.01374#page=6
• 8
  x.com
• 9
  Human average estimate (by Alan)