Selected historical models · estimates. Each vertical step is 100×.
What It Shows
Epoch’s estimates put the training computation for Llama 3.1-405B at about 80 million times that of AlexNet. That is a comparison of computation, not capability: these models were built for different tasks.
What It Does Not Show
Four selected historical models are not a complete trend or a ranking of today’s frontier. Epoch classifies all four records as “Confident”; its guidance still allows roughly a factor of three of uncertainty. This label covers the least certain populated quantity, not a separately measured error bar for each compute value.
Sources, Method & Data
Points use the dataset’s publication dates and training-compute estimates. The vertical axis is logarithmic: equal distances represent equal multipliers. No fitted trend line is shown. Selection and visualization by Vastkind.
Data: Epoch AI, Data on AI Models, accessed 5 October 2026. CC BY 4.0. Dataset Documentation · Original CSV.
| Model | Publication Date | Training Compute (FLOP) |
|---|---|---|
| AlexNet | 2012-09-30 | 4.7e+17 |
| GPT-3 175B | 2020-05-28 | 3.14e+23 |
| PaLM 540B | 2022-04-04 | 2.5272e+24 |
| Llama 3.1-405B | 2024-07-23 | 3.8e+25 |