Training compute growth
Maximum disclosed training compute per year on a log10(FLOP) scale. Missing values remain absent; the current year is incomplete.
Training compute growth
Total 468 · Page 12/19 · Observed 9/25/2026, 3:30:45 AM
| Name | Group | Reported value | Date | Notes | Source |
|---|---|---|---|---|---|
| Swin Transformer V2 (SwinV2-G) | Microsoft Research Asia | 1,100,000,000,000,000,000,000 FLOP | 2021-11-18 | Open weights (unrestricted) | Source ↗ |
| GPT-2 Medium (FlashAttention) | Stanford University,University at Buffalo | 892,809,220,000,000,000,000 FLOP | 2022-05-27 | Open weights (unrestricted) | Source ↗ |
| JFT | Google Research,Carnegie Mellon University (CMU) | 843,000,000,000,000,000,000 FLOP | 2017-07-10 | — | Source ↗ |
| CamemBERT | Facebook,INRIA,Sorbonne University | 830,000,000,000,000,000,000 FLOP | 2019-11-10 | Open weights (unrestricted) | Source ↗ |
| GenCast | Google DeepMind | 816,998,400,000,000,000,000 FLOP | 2024-05-01 | Open weights (non-commercial) | Source ↗ |
| Zidong Taichu | Chinese Academy of Sciences,Wuhan AI Computing Center | 801,600,000,000,000,000,000 FLOP | 2021-08-11 | Open weights (unrestricted) | Source ↗ |
| DD-PPO | Georgia Institute of Technology,Facebook AI Research,Oregon State University,Simon Fraser University | 780,000,000,000,000,000,000 FLOP | 2019-12-19 | Unreleased | Source ↗ |
| AlphaTensor | DeepMind | 714,147,840,000,000,000,000 FLOP | 2022-10-05 | Unreleased | Source ↗ |
| AlphaGo Zero | DeepMind | 649,439,910,290,227,000,000 FLOP | 2017-10-18 | Unreleased | Source ↗ |
| YOLOX-X | Megvii Inc | 634,275,000,000,000,000,000 FLOP | 2021-08-06 | Open weights (unrestricted) | Source ↗ |
| Once for All | MIT-IBM Watson AI Lab,Massachusetts Institute of Technology (MIT),IBM | 623,700,000,000,000,000,000 FLOP | 2020-04-29 | Open weights (unrestricted) | Source ↗ |
| OpenAI TI7 DOTA 1v1 | OpenAI | 604,609,522,259,200,200,000 FLOP | 2017-08-11 | — | Source ↗ |
| DiT-XL/2 | New York University (NYU),University of California (UC) Berkeley | 600,000,000,000,000,000,000 FLOP | 2023-03-02 | Open weights (non-commercial) | Source ↗ |
| Octo-Base | University of California (UC) Berkeley,Stanford University,Carnegie Mellon University (CMU),DeepMind | 585,000,000,000,000,000,000 FLOP | 2024-05-20 | Open weights (unrestricted) | Source ↗ |
| Libratus | Carnegie Mellon University (CMU) | 551,000,000,000,000,000,000 FLOP | 2017-08-19 | Unreleased | Source ↗ |
| Sparse all-MLP | Meta AI | 532,224,000,000,000,000,000 FLOP | 2022-04-14 | Unreleased | Source ↗ |
| AR-LDM | Alibaba,University of Waterloo,Vector Institute | 510,000,000,000,000,000,000 FLOP | 2022-11-20 | Unreleased | Source ↗ |
| Big Transformer for Back-Translation | Facebook AI Research,Google Brain | 478,080,000,000,000,000,000 FLOP | 2018-08-28 | Open weights (unrestricted) | Source ↗ |
| Masked Autoencoders ViT-H | Facebook AI Research | 460,000,000,000,000,000,000 FLOP | 2021-11-11 | Open weights (non-commercial) | Source ↗ |
| Xception | 436,000,000,000,000,000,000 FLOP | 2016-10-07 | — | Source ↗ | |
| Eurus-2-7B-PRIME | Tsinghua University,University of Illinois Urbana-Champaign (UIUC),Shanghai AI Lab,Peking University,Shanghai Jiao Tong University,CUHK Shenzhen Research Institute | 401,915,904,000,000,000,000 FLOP | 2025-01-02 | Open weights (unrestricted) | Source ↗ |
| DETR | 400,000,000,000,000,000,000 FLOP | 2020-05-26 | Open weights (unrestricted) | Source ↗ | |
| AlphaGo Fan | DeepMind | 380,000,000,000,000,000,000 FLOP | 2015-10-01 | Unreleased | Source ↗ |
| Transformer-XL Large + Phrase Induction | Massachusetts Institute of Technology (MIT),University of Illinois Urbana-Champaign (UIUC) | 378,486,510,000,000,000,000 FLOP | 2019-06-04 | Unreleased | Source ↗ |
| Transformer-XL (257M) | Carnegie Mellon University (CMU),Google Brain | 378,327,710,000,000,000,000 FLOP | 2019-01-09 | Open weights (unrestricted) | Source ↗ |
Data: Epoch AI · CC BY · View source · Automatically synced; not an exhaustive global sample