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 3/19 · Observed 9/25/2026, 3:30:45 AM
| Name | Group | Reported value | Date | Notes | Source |
|---|---|---|---|---|---|
| Kimi K2 Thinking | Moonshot | 4,200,000,000,000,000,000,000,000 FLOP | 2025-11-06 | Open weights (restricted use) | Source ↗ |
| Llama 4 Scout | Meta AI | 4,080,000,000,000,000,000,000,000 FLOP | 2025-04-05 | Open weights (restricted use) | Source ↗ |
| DeepSeek-R1 (May 2025) | DeepSeek | 4,020,010,000,000,001,000,000,000 FLOP | 2025-05-28 | Open weights (unrestricted) | Source ↗ |
| MegaScale (Production) | ByteDance,Peking University | 3,900,000,000,000,000,000,000,000 FLOP | 2024-02-23 | Unreleased | Source ↗ |
| Claude 2 | Anthropic | 3,866,000,000,000,000,000,000,000 FLOP | 2023-07-11 | API access | Source ↗ |
| EXAONE 4.5 | LG AI Research | 3,850,000,000,000,000,000,000,000 FLOP | 2026-04-09 | Open weights (restricted use) | — |
| Falcon-180B | Technology Innovation Institute | 3,760,000,000,000,000,000,000,000 FLOP | 2023-09-06 | Open weights (restricted use) | Source ↗ |
| LongCat-Flash | Meituan Inc | 3,726,000,000,000,000,000,000,000 FLOP | 2025-09-01 | Open weights (unrestricted) | Source ↗ |
| K-EXAONE 2.0 | LG AI Research | 3,550,000,000,000,000,000,000,000 FLOP | 2026-07-31 | Open weights (unrestricted) | Source ↗ |
| QwQ-32B | Alibaba | 3,510,000,000,000,000,000,000,000 FLOP | 2025-03-06 | Open weights (unrestricted) | Source ↗ |
| Qwen2.5-32B | Alibaba | 3,510,000,000,000,000,000,000,000 FLOP | 2024-09-17 | Open weights (unrestricted) | Source ↗ |
| DeepSeek-R1 | DeepSeek | 3,500,000,000,000,000,000,000,000 FLOP | 2025-01-20 | Open weights (unrestricted) | Source ↗ |
| Hunyuan-Large | Tencent | 3,492,370,000,000,000,000,000,000 FLOP | 2024-11-06 | Open weights (restricted use) | Source ↗ |
| DeepSeek-V3 (Mar 2025) | DeepSeek | 3,300,000,000,000,000,000,000,000 FLOP | 2025-03-24 | Open weights (restricted use) | Source ↗ |
| DeepSeek-V3 | DeepSeek | 3,300,000,000,000,000,000,000,000 FLOP | 2024-12-24 | Open weights (restricted use) | Source ↗ |
| GLM-5.3-Flash | Z.ai (Zhipu AI) | 3,240,000,000,000,000,000,000,000 FLOP | 2026-08-20 | Open weights (unrestricted) | Source ↗ |
| NVLM-X 72B | Nvidia | 3,039,818,100,000,000,000,000,000 FLOP | 2024-10-22 | Unreleased | Source ↗ |
| LLaVA-OV-72B | ByteDance,Nanyang Technological University,Chinese University of Hong Kong (CUHK),Hong Kong University of Science and Technology (HKUST) | 3,036,551,985,824,000,000,000,000 FLOP | 2024-08-06 | Open weights (unrestricted) | Source ↗ |
| NVLM-H 72B | Nvidia | 3,020,000,000,000,000,000,000,000 FLOP | 2024-10-22 | Unreleased | Source ↗ |
| NVLM-D 72B | Nvidia | 3,020,000,000,000,000,000,000,000 FLOP | 2024-10-22 | Open weights (non-commercial) | Source ↗ |
| Qwen2-72B | Alibaba | 3,020,000,000,000,000,000,000,000 FLOP | 2024-06-07 | Open weights (unrestricted) | Source ↗ |
| Kimi K2 | Moonshot | 2,976,000,000,000,000,000,000,000 FLOP | 2025-07-11 | Open weights (restricted use) | Source ↗ |
| Grok-1 | xAI | 2,900,000,000,010,000,000,000,000 FLOP | 2023-11-04 | Open weights (unrestricted) | Source ↗ |
| Minerva (540B) | 2,741,500,000,000,000,000,000,000 FLOP | 2022-06-29 | Unreleased | Source ↗ | |
| EXAONE 4.0 (32B) | LG AI Research | 2,690,000,000,000,010,000,000,000 FLOP | 2025-07-15 | Open weights (non-commercial) | Source ↗ |
Data: Epoch AI · CC BY · View source · Automatically synced; not an exhaustive global sample