OpenAI fulfills its promise, releases two new open-source AI models

OpenAI releases two new open-source AI models after five years, showcasing transparency and user control.

: OpenAI has introduced two new open-source AI models, named gpt-oss-120b and gpt-oss-20b, marking its first such release in five years. Aimed at restoring transparency, these models feature lightweight and flexible designs, with a focus on seamless software integration. The open-source nature allows users insight into the training parameters and enhances individual empowerment. CEO Sam Altman highlights this as a move toward user privacy and control, contrasting with OpenAI's traditionally closed model releases.

OpenAI, a prominent organization in artificial intelligence research and development, on August 5, 2025, released two open-source AI models: gpt-oss-120b and gpt-oss-20b. This move is significant as it's the first time in five years that OpenAI has introduced open-source models, aiming to return to a level of transparency not seen since prior to 2020. These models are specifically lightweight and flexible, designed to integrate easily into other software programs, supporting a variety of tasks in numerous settings. OpenAI's approach allows users to control and modify their AI directly when needed, a significant shift from the proprietary nature of many past OpenAI products.

The open-source initiative is part of a broader push towards transparency, helping users understand how AI systems function by disclosing training parameters. This disclosure provides a window into the reasons behind the models' behaviors, addressing long-standing criticisms about closed-off, proprietary AI solutions. Additionally, the open-source models fall under the Apache 2.0 license, allowing users to create new software without facing copyleft restrictions or patent risks. This initiative is expected to spur research and innovative product development, aligning with CEO Sam Altman's vision for individual empowerment.

In tandem with the open-source releases, OpenAI emphasized the importance of AI safety by embedding comprehensive safety training and evaluations into the models. The company further introduced an adversarially fine-tuned version of gpt-oss-120b, employing a preparedness framework to assess and track risky behavior in large language models. The renewed focus on safety underscores OpenAI's response to regulatory and ethical concerns that have grown in parallel with AI advancements.

The models are compatible with agentic workflows, which are processes driven by agent-based models, thereby expanding their applicability across different domains. By allowing local execution of these models, users gain the benefit of enhanced privacy while exploring new research opportunities. OpenAI's stance promotes an ecosystem where transparency and openness drive innovation, contrasting with its history of embracing closed ecosystems.

While OpenAI makes strides towards openness with these models, the future release of GPT-5, a highly anticipated and powerful model, is expected to be proprietary, reflecting ongoing debates about the organization's commitment to open strategies. This dual approach highlights the tension between fostering accessibility and maintaining control over highly sophisticated AI systems.

Sources: Gizmodo, OpenAI, X

AIOpen sourceMachine Learning