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 5/19 · Observed 9/25/2026, 3:30:45 AM
| Name | Group | Reported value | Date | Notes | Source |
|---|---|---|---|---|---|
| ESM3 (98B) | EvolutionaryScale,University of California (UC) Berkeley | 1,070,000,000,000,000,000,000,000 FLOP | 2024-06-25 | Unreleased | Source ↗ |
| ALLaM adapted 70B | Saudi Data and Artificial Intelligence Authority | 1,062,000,000,000,000,000,000,000 FLOP | 2024-05-21 | Unreleased | Source ↗ |
| Solar Open2 250B | Upstage | 1,050,000,000,000,000,000,000,000 FLOP | 2026-06-28 | Open weights (unrestricted) | Source ↗ |
| ERNIE 3.0 Titan | Baidu,Peng Cheng Laboratory | 1,042,100,000,000,000,000,000,000 FLOP | 2021-12-23 | Hosted access (no API) | Source ↗ |
| InternLM | Shanghai AI Lab,SenseTime | 998,400,000,000,000,000,000,000 FLOP | 2023-07-06 | Unreleased | Source ↗ |
| Megatron-Turing NLG 530B | Microsoft,Nvidia | 858,600,000,000,000,000,000,000 FLOP | 2021-10-11 | Unreleased | Source ↗ |
| Llama 2-70B | Meta AI | 810,000,000,000,000,000,000,000 FLOP | 2023-07-18 | Open weights (restricted use) | Source ↗ |
| Mixtral 8x7B | Mistral AI | 774,000,000,000,000,000,000,000 FLOP | 2023-12-11 | Open weights (unrestricted) | Source ↗ |
| AgentFounder-30B | Alibaba | 653,670,000,000,000,000,000,000 FLOP | 2025-09-16 | Open weights (unrestricted) | Source ↗ |
| Gopher (280B) | DeepMind | 631,000,000,000,000,000,000,000 FLOP | 2021-12-08 | Unreleased | Source ↗ |
| xTrimoPGLM -100B | Tsinghua University,BioMap Research | 620,000,000,000,000,000,000,000 FLOP | 2023-07-06 | Unreleased | Source ↗ |
| Yi-34B | 01.AI | 610,000,000,000,000,000,000,000 FLOP | 2023-11-02 | Open weights (restricted use) | Source ↗ |
| Llama 3.2 11B | Meta AI | 579,000,000,000,000,000,000,000 FLOP | 2024-09-24 | Open weights (restricted use) | Source ↗ |
| Chinchilla | DeepMind | 576,000,000,000,000,000,000,000 FLOP | 2022-03-29 | Unreleased | Source ↗ |
| PaLI-X | Google Research | 560,000,000,000,000,000,000,000 FLOP | 2023-05-29 | Unreleased | Source ↗ |
| LLaMA-65B | Meta AI | 550,000,000,000,000,000,000,000 FLOP | 2023-02-24 | Open weights (non-commercial) | Source ↗ |
| gpt-oss-20b | OpenAI | 549,000,000,000,000,000,000,000 FLOP | 2025-08-05 | Open weights (unrestricted) | Source ↗ |
| Parti | Google Research | 509,607,936,000,000,000,000,000 FLOP | 2022-06-22 | Unreleased | Source ↗ |
| MM1-30B | Apple | 486,000,000,000,000,000,000,000 FLOP | 2024-03-14 | Unreleased | Source ↗ |
| Nemotron 3-Nano-30B-A3B | Nvidia | 480,000,000,000,000,000,000,000 FLOP | 2025-12-15 | Open weights (restricted use) | Source ↗ |
| PanGu-Σ | Huawei Noah's Ark Lab | 467,000,000,000,000,000,000,000 FLOP | 2023-03-20 | Unreleased | Source ↗ |
| AFM-on-device | Apple | 451,260,000,000,000,000,000,000 FLOP | 2024-07-29 | Hosted access (no API) | Source ↗ |
| BlenderBot 3 | McGill University,Meta AI,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms) | 430,000,000,000,000,000,000,000 FLOP | 2022-08-10 | Open weights (non-commercial) | Source ↗ |
| OPT-175B | Meta AI | 430,000,000,000,000,000,000,000 FLOP | 2022-05-02 | Open weights (non-commercial) | Source ↗ |
| FunSearch | Google DeepMind | 387,000,000,000,000,000,000,000 FLOP | 2023-12-14 | Open weights (unrestricted) | Source ↗ |
Data: Epoch AI · CC BY · View source · Automatically synced; not an exhaustive global sample