According to insiders, OpenAI is preparing to launch its first open-weight model, which may be released as early as next week. This new language model will allow companies and governments to run it on vendors outside of OpenAI and Microsoft Azure servers, marking the first time since OpenAI released GPT-2 in 2019 that it has launched an open-weight model.
This move also breaks OpenAI's long-standing practice of keeping its language models closed-source since signing an exclusive cloud provider agreement with Microsoft in 2023.
OpenAI's models are usually closed-source, meaning the training parameters of the model are not made public. However, the upcoming open-weight model will change this situation, making it usable on Azure, Hugging Face, and other major cloud providers, similar to how DeepSeek's R1 model quickly gained support from Microsoft and other cloud providers earlier this year.
According to sources, the new model "is similar to o3mini" and has the reasoning capabilities that make OpenAI's latest models so powerful. Over the past few months, OpenAI has been demonstrating this open model to developers and researchers and actively seeking feedback from a broader AI community.
OpenAI has not yet commented on the upcoming open model. This move is undoubtedly going to attract widespread attention in the field of artificial intelligence and may have a profound impact on the future of model development and deployment.