OpenAI to Release First 'Open' Language Model Since GPT-2 in 2019

OpenAI is preparing to introduce an "open" version of its language model this year, which will provide developers the opportunity to run the model on their own hardware. This marks a significant shift from OpenAI's previous releases, such as GPT-3 and GPT-4, which have been closed systems.

On March 31, OpenAI CEO Sam Altman shared an update on X regarding the upcoming release of this powerful "new open-weight language model". Altman mentioned that while they are excited about the release, the company aims to gather feedback to ensure the model is as useful as possible for developers before its launch.

According to Altman, this model will be the first "open-weight" release from OpenAI since GPT-2 in 2019. He emphasized that they have been contemplating this direction for a while, highlighting the importance of this move at this point in time.

He further explained: "We still have some decisions to make, so we are hosting developer events to gather feedback and allow participants to experiment with early prototypes." These developer events are slated to take place in San Francisco, which is OpenAI's home base, with additional sessions planned for Europe and the Asia-Pacific region.

Altman expressed enthusiasm about observing how developers will use this model, particularly in cases where large companies and governments prefer to run their own models. He assured that consideration would be given to modifications that may occur post-release.

OpenAI’s First 'Open' Model Since GPT-2

The original open-weight language model, GPT-2, was initially released in February 2019, followed by a complete release in November of the same year. Altman also noted plans to ship GPT-4.5 and GPT-5 in the near future as competition in the AI landscape becomes increasingly fierce.

The AI market is rapidly evolving, with emerging competitors such as DeepSeek, which offers similar functionalities to ChatGPT but with potentially lower development costs. Notably, Alibaba has launched its own open-source AI model recently, while Google introduced its experimental Gemini 2.5 model.

Meanwhile, Meta's CEO remarked on social media that their AI model family, Llama, has achieved 1 billion downloads since its February 2023 release, highlighting the growing interest in AI models across the industry.

OpenAI, GPT-2, LanguageModel