OpenAI Releases Two New Open Source AI Models
OpenAI has released two new free and open-source artificial intelligence models, gpt-oss-120b and gpt-oss-20b, marking the company’s first open-source release in five years. Announced on Tuesday, August 5, 2025, these lightweight models are designed for easy integration into various software programs and are characterized by OpenAI as flexible yet powerful algorithms capable of performing diverse tasks across numerous settings.
The release signifies a notable shift for OpenAI, which has predominantly focused on proprietary models in recent years. OpenAI CEO Sam Altman expressed via X that he hopes these new AI models will facilitate “new kinds of research and the creation of new kinds of products.” He also emphasized the benefits of the open-source approach, stating, “We believe in individual empowerment. Although we believe most people will want to use a convenient service like ChatGPT, people should be able to directly control and modify their own AI when they need to, and the privacy benefits are obvious.”
A core advantage of open-source models, unlike most of OpenAI’s ChatGPT products, is the disclosure of the training parameters used to build their systems. This transparency allows developers and researchers to understand the models’ internal workings and behavior. The last time OpenAI released an open-source model was GPT-2, prior to the launch of ChatGPT in 2022 and when the company was still a relatively unknown startup.
The newly released models are free to use and can be run locally on users’ computers, eliminating the need to rely on OpenAI’s servers. They are specifically designed to integrate into “agentic workflows,” enabling new forms of automated tasks powered by AI agents. Furthermore, the models are released under the Apache 2.0 license, which permits users to create new software with the algorithms without concerns about copyleft restrictions or patent risks, whether for experimentation, customization, or commercial deployment. This licensing framework is crucial for fostering an open-source ecosystem where companies and individuals can build upon freely available models.
OpenAI also addressed AI safety in its announcement. The company stated that, in addition to conducting comprehensive safety training and evaluations, it implemented an “additional layer of evaluation by testing an adversarially fine-tuned version of gpt-oss-120b” using its Preparedness Framework. This framework is designed to assess and track potentially risky behavior in large language models.
This release comes amid past criticism directed at OpenAI for its “walled garden” approach, where its large language models remained proprietary and inaccessible for public inspection. By releasing these open-source models, OpenAI appears to be responding to calls for greater transparency and openness, aligning with its foundational name. However, the company’s long-term commitment to the Free and Open Source Software (FOSS) ethos remains a subject of observation, particularly given its substantial valuation. It is noteworthy that GPT-5, OpenAI’s highly anticipated and most powerful new model, is expected to follow the company’s recent pattern of closed, proprietary releases.