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 16/19 · Observed 9/25/2026, 3:30:45 AM
| Name | Group | Reported value | Date | Notes | Source |
|---|---|---|---|---|---|
| Feedback Transformer | LORIA,University of Lorraine,Facebook AI Research | 7,690,547,000,000,000,000 FLOP | 2020-02-21 | Unreleased | Source ↗ |
| Transformer | Google Research,Google Brain | 7,424,524,800,000,000,000 FLOP | 2017-06-12 | Unreleased | Source ↗ |
| HR-ResNet101 | Carnegie Mellon University (CMU) | 7,077,000,000,000,000,000 FLOP | 2016-12-13 | Open weights (unrestricted) | Source ↗ |
| ResNet-101 (ImageNet) | Microsoft | 7,004,000,000,000,000,000 FLOP | 2015-12-10 | Open weights (unrestricted) | Source ↗ |
| DLWP | University of Washington,Microsoft Research | 5,684,515,200,000,000,000 FLOP | 2021-02-09 | Unreleased | Source ↗ |
| Hanabi 4 player | DeepMind,University of Oxford,Carnegie Mellon University (CMU),Google Brain | 4,300,000,000,000,000,000 FLOP | 2019-02-01 | Unreleased | Source ↗ |
| DLRM-2020 | Facebook AI | 4,000,000,000,000,000,000 FLOP | 2019-05-31 | Unreleased | Source ↗ |
| AudioLM | Google Research | 3,900,000,000,000,000,000 FLOP | 2023-07-26 | Unreleased | Source ↗ |
| DeLighT | University of Washington,Allen Institute for AI,Facebook AI Research | 3,801,600,000,000,000,000 FLOP | 2020-08-03 | Unreleased | Source ↗ |
| BIDAF | University of Washington,Allen Institute for AI | 3,468,614,400,000,000,000 FLOP | 2016-11-05 | Open weights (unrestricted) | Source ↗ |
| SPPNet | Microsoft,Xi’an Jiaotong University,University of Science and Technology of China (USTC) | 3,411,072,000,000,000,000 FLOP | 2014-06-18 | — | Source ↗ |
| AWD-LSTM-MoS + dynamic evaluation (WT2, 2017) | Carnegie Mellon University (CMU) | 3,360,000,000,000,000,000 FLOP | 2017-11-10 | Unreleased | Source ↗ |
| DITTO | Tsinghua University,Apple,Westlake University,Chinese University of Hong Kong (CUHK) | 3,317,760,000,000,000,000 FLOP | 2022-06-06 | Unreleased | Source ↗ |
| TrellisNet | Carnegie Mellon University (CMU),Bosch Center for Artificial Intelligence,Intel Labs | 2,780,000,000,000,000,000 FLOP | 2018-10-15 | Unreleased | Source ↗ |
| PermuteFormer | Peking University | 2,775,000,000,000,000,000 FLOP | 2021-09-06 | Unreleased | Source ↗ |
| DistBelief NNLM | 2,612,736,000,000,000,000 FLOP | 2013-01-16 | — | Source ↗ | |
| Cross-lingual alignment | Tel Aviv University,Massachusetts Institute of Technology (MIT) | 2,560,000,000,000,000,000 FLOP | 2019-04-04 | Open weights (unrestricted) | Source ↗ |
| RoFormer | Zhuiyi Technology | 2,162,688,000,000,000,000 FLOP | 2023-11-08 | Unreleased | Source ↗ |
| DensePhrases | Korea University,Princeton University | 2,099,520,000,000,000,000 FLOP | 2020-12-23 | Open weights (unrestricted) | Source ↗ |
| RetinaNet-R101 | Facebook AI Research | 2,065,392,000,000,000,000 FLOP | 2017-08-07 | — | Source ↗ |
| Transformer-XL DeFINE (141M) | University of Washington,Allen Institute for AI | 1,742,760,000,000,000,000 FLOP | 2019-11-27 | Unreleased | Source ↗ |
| Tensor-Transformer(1core)+PN (WT103) | University of California (UC) Berkeley | 1,580,000,000,000,000,000 FLOP | 2020-03-17 | Open weights (unrestricted) | Source ↗ |
| RNNsearch-50* | Jacobs University Bremen,University of Montreal / Université de Montréal | 1,555,200,000,000,000,000 FLOP | 2014-09-01 | — | Source ↗ |
| GoogLeNet / InceptionV1 | Google,University of Michigan,University of North Carolina | 1,510,000,000,000,000,000 FLOP | 2014-09-17 | — | Source ↗ |
| TransE | Universite de Technologie de Compiègne – CNRS,Google | 1,340,928,000,000,000,000 FLOP | 2013-12-05 | — | Source ↗ |
Data: Epoch AI · CC BY · View source · Automatically synced; not an exhaustive global sample