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 10/19 · Observed 9/25/2026, 3:30:45 AM
| Name | Group | Reported value | Date | Notes | Source |
|---|---|---|---|---|---|
| ResNeXt-101 32x48d | 8,743,950,000,000,000,000,000 FLOP | 2018-05-02 | Open weights (non-commercial) | Source ↗ | |
| RoBERTa Large | Facebook,University of Washington | 8,506,700,000,000,000,000,000 FLOP | 2019-07-01 | Open weights (unrestricted) | Source ↗ |
| ContextNet + Noisy Student | 8,160,000,000,000,000,000,000 FLOP | 2020-01-19 | Unreleased | Source ↗ | |
| Nucleotide Transformer | Nvidia,Technical University of Munich,InstaDeep | 8,080,000,000,000,000,000,000 FLOP | 2023-01-15 | Open weights (non-commercial) | Source ↗ |
| Segment Anything Model | Meta AI | 7,800,000,000,000,000,000,000 FLOP | 2023-04-05 | Open weights (unrestricted) | Source ↗ |
| CHAI-1 | Chai discovery | 7,760,572,400,000,000,000,000 FLOP | 2024-10-15 | Open weights (non-commercial) | Source ↗ |
| Conformer + Wav2vec 2.0 + Noisy Student | Google,Google Research,Google Brain | 7,600,000,000,000,000,000,000 FLOP | 2020-10-20 | Unreleased | Source ↗ |
| GGNN | Westlake University,Tsinghua University,Toyota Technological Institute at Chicago | 7,560,000,000,000,000,000,000 FLOP | 2023-08-05 | Unreleased | Source ↗ |
| DINOv2 | Facebook AI Research,INRIA | 7,418,511,360,000,000,000,000 FLOP | 2023-04-14 | Open weights (unrestricted) | Source ↗ |
| NÜWA | Microsoft Research,Peking University | 7,245,987,840,000,000,000,000 FLOP | 2021-11-24 | Unreleased | Source ↗ |
| Tranception | University of Oxford,Harvard Medical School,Cohere | 7,240,000,000,000,000,000,000 FLOP | 2022-05-27 | Open weights (unrestricted) | Source ↗ |
| GNMT | 6,620,000,000,001,000,000,000 FLOP | 2016-09-26 | Hosted access (no API) | Source ↗ | |
| Ankh_large | Technical University of Munich,Columbia University | 6,500,000,000,000,000,000,000 FLOP | 2023-01-16 | Open weights (non-commercial) | Source ↗ |
| Hybrid H3-2.7B | Stanford University,University at Buffalo | 6,480,000,000,000,000,000,000 FLOP | 2022-12-28 | Open weights (unrestricted) | Source ↗ |
| Make-A-Scene | Meta AI | 6,417,285,100,000,000,000,000 FLOP | 2022-03-24 | Unreleased | Source ↗ |
| ADM | OpenAI | 6,200,000,000,000,000,000,000 FLOP | 2021-05-11 | Open weights (non-commercial) | Source ↗ |
| XLNet | Carnegie Mellon University (CMU),Google Brain | 6,190,000,000,000,000,000,000 FLOP | 2019-06-01 | Open weights (unrestricted) | Source ↗ |
| HuBERT | Facebook AI Research | 5,540,000,000,000,000,000,000 FLOP | 2021-07-27 | Open weights (unrestricted) | Source ↗ |
| M6-T | Alibaba | 5,500,000,000,000,000,000,000 FLOP | 2021-03-05 | Unreleased | Source ↗ |
| MSA Transformer | Facebook AI Research,University of California (UC) Berkeley,New York University (NYU) | 5,490,000,000,000,000,000,000 FLOP | 2021-02-13 | Open weights (unrestricted) | Source ↗ |
| ESM1b | Facebook AI Research,New York University (NYU) | 5,100,000,000,000,000,000,000 FLOP | 2020-12-15 | Open weights (unrestricted) | Source ↗ |
| SigLIP 400M | Google DeepMind | 4,946,730,100,000,000,000,000 FLOP | 2023-03-27 | Open weights (unrestricted) | Source ↗ |
| AlphaFold-Multimer | Google DeepMind,DeepMind | 4,350,000,000,000,000,000,000 FLOP | 2021-10-04 | Open weights (unrestricted) | Source ↗ |
| ViT-Huge/14 | Google Brain,Google Research | 4,262,000,000,000,000,000,000 FLOP | 2020-10-22 | Open weights (unrestricted) | Source ↗ |
| Whisper | OpenAI | 4,207,266,300,000,000,000,000 FLOP | 2022-09-21 | Open weights (unrestricted) | Source ↗ |
Data: Epoch AI · CC BY · View source · Automatically synced; not an exhaustive global sample