Artificial Intelligence pioneer, OpenAI, fulfills its promise, and unveils two new AI models as open-source creations
OpenAI Embraces Open-Source AI with the Release of gpt-oss-120b and gpt-oss-20b
OpenAI, the billion-dollar organisation known for its advanced AI models, has taken a significant step towards openness with the recent release of gpt-oss-120b and gpt-oss-20b. These open-weight AI reasoning models are now available under the permissive Apache 2.0 license, allowing enterprises to monetize and build upon them without seeking permission from OpenAI.
The new models are designed to integrate into agentic workflows, enabling new types of automated work by so-called "agents." OpenAI characterises them as flexible and powerful, capable of performing various tasks and being used in numerous settings.
In a statement, OpenAI's CEO, Sam Altman, highlighted the potential of these models to assist with new kinds of research and product creation. He also advocated for individual empowerment and the privacy benefits of open-source AI.
However, OpenAI's commitment to openness does not extend to the training datasets for these models. Citing ongoing legal and ethical issues related to copyrighted material in AI training data, the company has chosen not to release these datasets. This decision is aimed at addressing safety concerns, as malicious actors could potentially modify the models to bypass safety measures. Developers using the open models will therefore need to implement their own safeguards tailored to their systems.
OpenAI has run the models through comprehensive safety training and evaluations, and has also introduced an additional layer of evaluation using its Preparedness Framework. To further explain the models and how they work, the company has launched a feedback portal and a more extensive blog.
Users can run the models locally on their computers, provided their computer specifications meet the requirements. The models are lightweight and designed for easy integration into other software programs.
This move towards open-source AI marks a shift in OpenAI's approach, as the company has historically been criticised for its "walled garden" approach to software development. However, OpenAI is now trying to prove the haters wrong and double down on its commitment to being a truly "open" organisation.
The company frames the choice between open and closed source AI as a false dichotomy, advocating for a complementary ecosystem where both open models and proprietary models coexist to foster innovation, safety, and democratic control over AI technology. The open models are intended to empower governments, institutions, and emerging markets where data residency or security requirements make fully cloud-hosted AI unfeasible.
While the new models are a significant step towards openness, it is worth noting that OpenAI's most powerful and highly anticipated new model, GPT-5, is expected to be released in the same fashion as other recent GPT releases: closed.
[1] OpenAI Blog: [Link to the blog post] [2] TechCrunch: [Link to the TechCrunch article] [3] Wired: [Link to the Wired article] [4] VentureBeat: [Link to the VentureBeat article]
- The recently released gpt-oss-120b and gpt-oss-20b by OpenAI are open-weight AI reasoning models, which not only cater to the tech industry but also AI enthusiasts interested in artificial-intelligence and its future development.
- OpenAI's commitment to open-source AI extends beyond just releasing models; they have launched a feedback portal and an extensive blog to provide better insights into their technology and its applications in tech, artificial-intelligence, and beyond.
- Despite the significant step towards openness with gpt-oss-120b and gpt-oss-20b, OpenAI has clarified that its most powerful model, GPT-5, will be released in the traditional closed manner, emphasizing the coexistence of open and proprietary models in fostering innovation, safety, and democratic control over AI technology.
