OpenAI launches GPT-Oss: two new open (and powerful) models that you can download on your computer
OpenAI has announced the launch of GPT-Oss-120B and GPT-Oss-20Btwo language models of open source that mark a significant advance in the reasoning capacity of this type of models. Available under the flexible Apache 2.0 license, according to the company, stand out for its higher performance to other open models of similar size in reasoning tasks, its tools use capabilities and optimization for efficient implementation in consumption hardware.
These are OpenAi's first open models from GPT 2, launched in 2020.
As explained Sam Altman During the presentation: «We are excited to put this model, the result of billions of dollars in research, within the reach of the world, so that AI reaches the greatest possible number of people. With this launch, we trust that new lines of research are enabled and new types of products are created. (…) We are excited that the world can build on an open AI infrastructure, created in the United States, based on democratic values, available for free and oriented to collective benefit ».
Openai ensures that, given their size, these models suppose significant advances both in reasoning and security and that their objective by releasing them is to accelerate avant -garde research and allow a more sure and transparent development of AI. The idea is that these Open Models also help reduce Barriers for emerging marketsthe sectors with limited resources and the smallest organizations that perhaps do not have the necessary budget or flexibility to adopt proprietary models.
How are OpenAi's new open models
The GPT-Oss-120b model almost equals the OPENAI O4-MINI model in central reasoning tests and can work efficiently in a single 80 GB GPU. For its part, GPT-Oss-20B offers similar results to those of OpenAi O3-MINI in common evaluations and It only requires 16 GB of memory. Both models admit a context length of up to 128K and were trained with a high quality text data set, focused on STEM, programming and general knowledge.
Both models have been trained with a combination of reinforcement learning and advanced techniques inspired by Openai's internal models. In evaluations, GPT-Oss-120b surpasses O3-mini And, in some cases, equals or exceeds O4-mini in areas such as competitive programming, general problems resolving, the use of tools and medical care.
Openai open models performance. Source: OpenAi
On the other hand, security is a fundamental component in the OpenAI approach for these releases. The models underwent exhaustive safety training and evaluation, and an adverse adjusted version of GPT-Oss-120b was tested to evaluate the risks of malicious use. As part of this initiative, Openai will also organize a “Red Teaming” challenge with A prize of $ 500,000 for the community to identify new security vulnerabilities.
The files for both models are Available for free download in Hugging Face.
Image: OpenAi
