OpenAI has announced the launch of GPT-OSS, a new open-weight artificial intelligence model now available free to developers, researchers, and companies worldwide.
The release includes two versions: a large model with 120 billion parameters and a smaller 20 billion-parameter model that can run on a laptop with 16GB RAM. The bigger model requires more powerful hardware such as an 80GB GPU.
These models are designed to perform complex tasks and mimic human reasoning. They are accessible on popular platforms like Hugging Face, Amazon Bedrock, Groq’s inference cloud, and Saudi Arabia’s Humain AI. This launch marks OpenAI’s first open-weight model since 2019’s GPT-2, allowing developers to access and build upon the models’ underlying parameters, known as weights. However, OpenAI has not made the training data publicly available, so the release falls short of being fully open source.
OpenAI CEO Sam Altman said that the company delayed the launch to conduct extra safety reviews after concerns about open-weight AI security risks. He noted the challenge that once weights are released, they cannot be taken back, a first for OpenAI. Despite this, the company anticipates users, including companies such as Orange SA and Snowflake Inc., will experiment and provide feedback to shape future model releases.
Chinese firm DeepSeek and Meta with its LLaMA models have also contributed to the open AI ecosystem.
OpenAI offers resources including an open model playground and guides for developers to get started with the new models. The ongoing dialogue between AI creators and users will likely influence the next chapter in open AI technology worldwide.
