OpenAI is preparing to release its first open-weighted model, following a long history of closed-system language models. The model is expected to be available next week. This development will mark the first time the company has released a language model that shares its weighting parameters with users since GPT-2, which was announced in 2019.
OpenAI is preparing to release its open AI model
The new model will not be limited to Microsoft Azure. It can also be run on other major cloud providers, such as Hugging Face. This makes OpenAI models available on infrastructures outside of Azure for the first time. It is stated that companies and public institutions will be able to run the model by integrating it into their own servers.
The model is said to share technical features with the “o3 mini.” It is known as one of OpenAI’s recently developed small-scale models with powerful reasoning and understanding capabilities. The company demoed the new open-weighted model to a limited group of developers and researchers in recent months, gathering feedback. However, no official launch has been made so far.
The new model stands out as one of the first significant changes following the collaboration agreement signed between Microsoft and OpenAI in 2023. This agreement granted Microsoft priority access and commercial use of OpenAI models. This privilege is now broken. The fact that competing providers also have access to the same model is seen as a step that could shift the balance in the AI industry.
OpenAI describes this work as an “open model.” However, this should not be confused with open source. The model’s weightings will be publicly available, but users will not be able to modify them. The company is also expected to partially disclose its training data.
With the release of the model, OpenAI’s move from a closed ecosystem to a more transparent structure will both increase access for researchers and pave the way for AI solutions that can be used in public infrastructures.
{{user}} {{datetime}}
{{text}}