Recently, the complicated tie between Microsoft and OpenAI has taken a new turn. Since the entities continue to renegotiate a deal to allow OpenAI to restructure into a profitable organization. OpenAI is looking to release an OpenAI’s open language model, which could contribute more than a bridge between the two entities. Let’s explore more about this.
What is the Buzz Around OpenAI’s Open Language Model?
Reportedly, the AI lab of Sam Altman is prepping an open-weight model which will roll out next week with providers other than just OpenAI and Microsoft’s Azure servers. OpenAI’s open language model is generally closed-weight. This implies that the weights are not publicly accessible.
The open nature of OpenAI’s open language model means the organizations and governments will run the model themselves, similar to the way Microsoft and other cloud suppliers quickly adopted DeepSeek’s R1 model at the beginning of 2025.
It is expected to be available on Azure, Hugging Face, and other major cloud providers soon. Sources describe the model as similar to the O3 mini, full of reasoning capabilities that make OpenAI’s newest models more powerful. OpenAI has been presenting this open model to researchers and developers recently and has been actively collecting feedback from the wider AI community.
This is the first time OpenAI has launched an open-weight model since introducing GPT-2 in 2019. Additionally, it is the first time that OpenAI’s open language model has been launched since its agreement with Microsoft as an exclusive cloud provider in 2023. This agreement means most of OpenAI’s models are accessed by Microsoft. Microsoft also has the exclusive rights to sell them directly to organizations through its own Azure OpenAI services. However, there is nothing to compete with competitors in cloud operations with an open model, not even hosting a version of it.
Agreement Between OpenAI and Microsoft
As noted in Notepad, there is an intricate revenue-sharing arrangement between OpenAI and Microsoft, which allocates 20% of the revenue from ChatGPT and the AI startup’s API platform to Microsoft. The tech giant also shares 20% of the Azure OpenAI revenue directly with OpenAI. OpenAI’s open language model may have an impact on the own AI business of Microsoft. The open model may imply that some Azure users will not require costly options, or they may even shift to competing cloud providers.
The attractive exclusivity offering from Microsoft, in collaboration with OpenAI, has been in trial for the past few months. Microsoft changed its OpenAI agreement earlier this year to enable the AI lab to get its own AI compute from competitors like Oracle. Although it was initially restricted to the servers used for developing AI models, this new OpenAI’s open language model will surpass the boundaries of ChatGPT and Azure OpenAI. Though Microsoft has the right to refuse to supply computing resources to OpenAI, it cannot control the open language model.
OpenAI is planning to announce the language model as ‘open model’. However, the term that is often confused with open-source is prone to debate about how open the new model is. That will depend on the attached license and whether OpenAI wishes to offer full access to the model’s code and training information, which other scholars can then copy.
When Will the New Model be Release?
Altman opined that this open-weight language model may arrive in the upcoming months. This could be released next week. However, OpenAI’s release dates often change, which increases uncertainties related to development issues, server capacity, competitor AI launches, and even leaks. Nevertheless, we can expect the launch of an open language model this month, provided everything remains on schedule.
What’s More?
O3, an AI model developed by the creators of ChatGPT, has positioned itself as the top AI tool for addressing scientific queries across various fields, according to a benchmarking platform launched last week.
SciArena, designed by the Allen Institute for AI in Seattle, Washington, was ranked 23rd among large language models based on their responses to scientific queries. Around 102 researchers cite the quality of the answers. OpenAI’s o3 was recognized for providing the best answers to questions in the natural sciences, engineering, healthcare, humanities, and social sciences, receiving approximately 13,000 votes.
DeepSeek-R1, designed by DeepSeek in Hangzhou, China, ranked second on natural-sciences queries and fourth on engineering. The third position is acquired by Google’s Gemini 2.5 Pro in natural sciences, and the fifth in engineering and healthcare.
The user’s inclination towards O3 could be due to the model’s tendency to provide sufficient detail on the literature it references and to generate technically sound responses. However, explaining the variance in the performance of the mode is a daunting task because most are proprietary. The variances in the training data and the optimisation of the data could partially help explain this.
SciArena is the newest platform designed to examine how AI models perform on specific tasks and is one of the first to rank performance on scientific tasks using crowdsourced feedback. A robotics and AI researcher opined that SciArena is a positive attempt that inspires a thoughtful evaluation of LLM-assisted literature tasks.
Also Read:
Siri to be Now Powered by OpenAI? Unveiling the Truth
OpenAI Rolls Out Canvas: Revolutionizing AI-Assisted Workspaces
