OpenAI releases open-weight version that can run on laptops

OpenAI says it has released two open weight language models that excel at advanced inference and are optimized to run on laptops with similar performance levels to its proprietary 4o-mini inference models.

The trained parameters or weights of an open weighted language model are publicly accessible, used by developers to analyze and fine-tune the model for specific tasks without needing the original training data.

“One of the unique things about open models is that people can run them locally. People can run them behind their own firewalls, on their own infrastructure,” OpenAI co-founder Greg Brockman said in a press release.

The 120B and 20B gpt-oss versions can run natively on laptops.

Open weight language models differ from open source models in that they provide access to the entire source code, training data, and methodology.

Separately, Amazon announced that OpenAI’s open weight models are now available on Amazon Web Services’ Bedrock Generative AI marketplace. This is the first time an OpenAI model has been made available on Bedrock, said Atul Deo, product manager at Bedrock.

“OpenAI has been developing great models, and we believe that these models will be great open source options, or open modeling options for customers,” Deo said in an interview. He declined to discuss any contractual arrangements between AWS and OpenAI.

Amazon last week after the company reported slowing growth in its AWS unit, especially when compared to competitors.

The gpt-oss-20b model is sold on the Sage Maker e-commerce platform.

The landscape of open-source AI models and open weights has been a contentious one this year. For a while, META’s Llama models were considered the best, but that changed earlier this year when China’s DeepSeek released a powerful and cost-effective reasoning model, while Meta struggled to develop Llama 4.

The two new OpenAI models are the first open models OpenAI has released since GPT-2, which was released in 2019.

OpenAI's larger model, gpt-oss-120b, can run on a single GPU, and the second model, gpt-oss-20b, is small enough to run directly on a personal computer, the company said.

OpenAI says these models perform similarly to its proprietary inference models called o3-mini and o4-mini, and particularly excel at programming, competitive math, and health-related queries.

The models are trained on a text-only dataset that focuses on science , math, and programming in addition to general knowledge. OpenAI does not publish benchmarks comparing its open-weight models to competing models like DeepSeek-R1.

Microsoft-backed OpenAI, currently valued at $300 billion, is now raising up to $40 billion in a new funding round led by Softbank Group.