OpenAI’s GPT-2 which was released in 2019 is still one of the most standout large language models and was downloaded 15.7 million times on HuggingFace repository in the past month. There have been many more advanced LLMs right now and as compared to them, GPT-2’s data is limited and only contains web pages from Reddit’s outbound links with 3 karma at least. The model’s page on HuggingFace also says that GPT-2 doesn't have Wikipedia content. AI models by Meta are also included in the link but none of them are latest and this shows that other companies are releasing more advanced models unlike GPT-2 who has maintained user interest even with limited capabilities.
On the other hand, Meta released OPT-125M in 2022 which is listed under Facebook’s name on HuggingFace and had 6.13 million downloads during the past month which shows that some users are still interested in this model. Llama 3.1 has 5.71 million downloads, making it third most downloaded model on HuggingFace while Llama 3.3 had 603,000 downloads in the past month. DeepSeek-R1, the newcomer, recorded 674,452 downloads last month.
Nemo Instruct Model by MistralAI was downloaded 1.46 million times while OpenELM 1.1B Instruct Model by Apple got 1.39 million downloads in the past month. Unlike apps like ChatGPT, open-source models can be downloaded and used by everyone as long as they have a license. But free models like Llama don't have clarity about what data they were trained on and using them can be tricky.
GPT-2's impressive download numbers suggest that older AI models still hold significant appeal, even in a market flooded with newer, more advanced alternatives. Despite its limitations, such as a restricted data set and lack of Wikipedia content, GPT-2 continues to attract users with its accessibility and reliability. This enduring popularity highlights the value many place on tried-and-true models, as well as the demand for open-source solutions that are easy to deploy. In a landscape where cutting-edge AI models dominate the headlines, GPT-2’s continued success may signal that older models, with their established frameworks, still have a vital place in the AI ecosystem.
Read next: Global AI Safety Report Warns of Cyber Threats, Manipulation, and Weaponization Risks
On the other hand, Meta released OPT-125M in 2022 which is listed under Facebook’s name on HuggingFace and had 6.13 million downloads during the past month which shows that some users are still interested in this model. Llama 3.1 has 5.71 million downloads, making it third most downloaded model on HuggingFace while Llama 3.3 had 603,000 downloads in the past month. DeepSeek-R1, the newcomer, recorded 674,452 downloads last month.
Nemo Instruct Model by MistralAI was downloaded 1.46 million times while OpenELM 1.1B Instruct Model by Apple got 1.39 million downloads in the past month. Unlike apps like ChatGPT, open-source models can be downloaded and used by everyone as long as they have a license. But free models like Llama don't have clarity about what data they were trained on and using them can be tricky.
GPT-2's impressive download numbers suggest that older AI models still hold significant appeal, even in a market flooded with newer, more advanced alternatives. Despite its limitations, such as a restricted data set and lack of Wikipedia content, GPT-2 continues to attract users with its accessibility and reliability. This enduring popularity highlights the value many place on tried-and-true models, as well as the demand for open-source solutions that are easy to deploy. In a landscape where cutting-edge AI models dominate the headlines, GPT-2’s continued success may signal that older models, with their established frameworks, still have a vital place in the AI ecosystem.
Read next: Global AI Safety Report Warns of Cyber Threats, Manipulation, and Weaponization Risks
