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 9/19 · Observed 9/25/2026, 3:30:45 AM
| Name | Group | Reported value | Date | Notes | Source |
|---|---|---|---|---|---|
| ERNIE 3.0 | Baidu | 22,500,000,000,000,000,000,000 FLOP | 2021-07-05 | Open weights (unrestricted) | Source ↗ |
| Megatron-BERT | Nvidia | 22,000,000,000,000,000,000,000 FLOP | 2019-09-17 | Unreleased | Source ↗ |
| GraphCast | Google DeepMind | 21,000,000,000,000,002,000,000 FLOP | 2023-11-14 | Open weights (non-commercial) | Source ↗ |
| XLM-RoBERTa | Facebook AI | 20,760,000,000,000,000,000,000 FLOP | 2019-11-05 | Open weights (non-commercial) | Source ↗ |
| ProtT5-XL-U50 | Technical University of Munich,Med AI Technology,Nvidia,Oak Ridge National Laboratory,Google,Seoul National University | 18,704,498,688,000,000,000,000 FLOP | 2021-05-04 | Open weights (non-commercial) | Source ↗ |
| LUKE | University of Washington,National Institute of Informatics | 18,144,000,000,000,000,000,000 FLOP | 2020-10-02 | Open weights (unrestricted) | Source ↗ |
| SEER | Facebook AI Research,INRIA | 18,000,000,000,000,000,000,000 FLOP | 2021-07-29 | Open weights (non-commercial) | Source ↗ |
| Statement Curriculum Learning | OpenAI | 17,901,648,000,000,000,000,000 FLOP | 2022-03-02 | API access | Source ↗ |
| NLLB | Meta AI | 17,511,137,280,000,000,000,000 FLOP | 2022-07-06 | Open weights (unrestricted) | Source ↗ |
| RETRO-7B | DeepMind | 16,800,000,000,000,000,000,000 FLOP | 2022-02-07 | Unreleased | Source ↗ |
| Turing-NLG | Microsoft | 15,700,000,000,000,000,000,000 FLOP | 2020-02-13 | Unreleased | Source ↗ |
| EVA-01 | Beijing Academy of Artificial Intelligence / BAAI,Huazhong University of Science and Technology,Zhejiang University (ZJU),Beijing Institute of Technology | 15,010,000,000,000,000,000,000 FLOP | 2022-11-14 | Open weights (unrestricted) | Source ↗ |
| Imagen | Google Brain | 14,600,000,000,000,000,000,000 FLOP | 2022-05-23 | API access | Source ↗ |
| Generative BST | Facebook AI Research | 14,490,000,000,000,000,000,000 FLOP | 2021-03-05 | Open weights (unrestricted) | Source ↗ |
| ProGen2-xlarge | Salesforce Research,Columbia University,Johns Hopkins University | 13,500,000,000,000,000,000,000 FLOP | 2022-06-27 | Open weights (unrestricted) | Source ↗ |
| OpenAI Five Rerun | OpenAI | 13,000,000,000,000,000,000,000 FLOP | 2019-12-13 | Unreleased | Source ↗ |
| MolmoAct 2 | Allen Institute for AI,University of Washington,National University of Singapore,University of Pennsylvania,Johns Hopkins University,Amazon,Cortex AI | 11,000,000,000,000,000,000,000 FLOP | 2026-04-05 | Open weights (unrestricted) | Source ↗ |
| CLIP (ViT L/14@336px) | OpenAI | 10,500,000,000,000,000,000,000 FLOP | 2021-01-05 | Open weights (unrestricted) | Source ↗ |
| OmegaPLM | Massachusetts Institute of Technology (MIT),Westlake University | 10,351,411,200,000,000,000,000 FLOP | 2022-07-22 | Unreleased | Source ↗ |
| PLATO-XL | Baidu | 9,900,000,000,000,000,000,000 FLOP | 2021-09-20 | Open weights (unrestricted) | Source ↗ |
| VideoMAE V2 | Nanjing University,Shenzhen Institute of Advanced Technology,Shanghai AI Lab | 9,700,000,000,000,000,000,000 FLOP | 2023-03-29 | Open weights (unrestricted) | Source ↗ |
| FGN | Google DeepMind | 9,618,950,880,000,001,000,000 FLOP | 2025-06-12 | Unreleased | Source ↗ |
| AudioGen | Meta AI,Hebrew University of Jerusalem | 9,500,000,000,000,000,000,000 FLOP | 2023-03-05 | Open weights (non-commercial) | Source ↗ |
| Megatron-LM (8.3B) | Nvidia | 9,100,000,000,000,000,000,000 FLOP | 2019-09-17 | Unreleased | Source ↗ |
| T5-3B | 9,000,000,000,100,000,000,000 FLOP | 2019-10-23 | Open weights (unrestricted) | Source ↗ |
Data: Epoch AI · CC BY · View source · Automatically synced; not an exhaustive global sample