Frontier language models from United States of America
38
Models
17 organisations
29%
Open weights
9 open, 22 closed
38
Frontier
By Epoch's flag
1.6T
Largest traceable
Switch
1959–2025
Span
20 with a traceable size
Organisation
Records
38| Model | Organisation | Released | Parameters | Weights |
|---|---|---|---|---|
| Grok 4 | xAI | 2025-07 | 3.0T * | Closed |
| GPT-4.5 | OpenAI | 2025-02 | — | Closed |
| Grok 3 | xAI | 2025-02 | 3.0T * | Closed |
| Grok-2 | xAI | 2024-08 | — | Closed |
| Llama 3.1-405B | Meta AI | 2024-07 | 405B | Open |
| Nemotron-4 340B | NVIDIA | 2024-06 | 340B | Open |
| Gemini 1.5 Pro | Google DeepMind | 2024-02 | — | Closed |
| Inflection-2 | Inflection AI | 2023-11 | — | Closed |
| Amazon Titan | Amazon | 2023-09 | 200B * | Closed |
| Claude 2 | Anthropic | 2023-07 | — | Closed |
| PaLM 2 | 2023-05 | 340B * | Closed | |
| Flan-PaLM 540B | 2022-10 | 540B | Closed | |
| U-PaLM (540B) | 2022-10 | 540B | Closed | |
| Minerva (540B) | 2022-06 | 540B | Closed | |
| PaLM (540B) | Google Research | 2022-04 | 540B | Closed |
| GPT-3.5 (davinci-002) | OpenAI | 2022-03 | — | Closed |
| Megatron-Turing NLG 530B | Microsoft | 2021-10 | 530B | Closed |
| FLAN 137B | Google Research | 2021-09 | 137B | Closed |
| Switch | 2021-01 | 1.6T | Open | |
| mT5-XXL | 2020-10 | 13B | Open | |
| GShard (dense) | 2020-06 | 2B | Closed | |
| GPT-3 175B (davinci) | OpenAI | 2020-05 | 175B | Closed |
| Meena | Google Brain | 2020-01 | 3B | Closed |
| XLM-RoBERTa | Facebook AI | 2019-11 | 550M | Open |
| T5-11B | 2019-10 | 11B | Open | |
| T5-3B | 2019-10 | 3B | Open | |
| Megatron-BERT | NVIDIA | 2019-09 | 4B | Closed |
| Megatron-LM (8.3B) | NVIDIA | 2019-09 | 8B * | Closed |
| RoBERTa Large | 2019-07 | 355M | Open | |
| XLNet | Carnegie Mellon University (CMU) | 2019-06 | 340M | Open |
| GNMT | 2016-09 | 278M * | Closed | |
| SNM-skip | 2014-12 | 62B * | Not recorded | |
| Seq2Seq LSTM | 2014-09 | 2B | Not recorded | |
| DistBelief NNLM | 2013-01 | — | Not recorded | |
| RNN LM | Johns Hopkins University | 2010-09 | 70M * | Not recorded |
| KN-LM | 2007-06 | 21B * | Not recorded | |
| SB-LM | 2007-06 | 300B * | Not recorded | |
| Pandemonium (morse) | Massachusetts Institute of Technology (MIT) | 1959-02 | — | Not recorded |
An asterisk marks a parameter figure the source does not consider traceable to the people who built the model. That judgement is the source’s and it has false negatives: Kimi K3 is marked speculative at 2.8T while Moonshot’s own model card states the figure. Why that distinction carries the whole exhibit.
Counts are of models Epoch AI records as notable, not of every model released. The most recent month is always incomplete, because a model enters the dataset when it is reviewed rather than when it ships.
Epoch AI, 'Data on AI Models'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/data/ai-models-documentation' Used under CC BY 4.0.