In a major shift in strategy, OpenAI has released two new open-weight AI language models — gpt-oss-120B and gpt-oss-20B — marking the company’s return to open AI development after six years. These models are designed to be run on personal hardware such as laptops and can be fine-tuned for various use cases by developers, businesses, and even governments.
OpenAI last launched an open-weight model in 2019 with GPT-2. Since then, it has focused on proprietary, closed-source models like GPT-3, GPT-4, and the recently released GPT-4o. The move to re-enter the open model space seems to have been accelerated by the rise of competitive open-weight models, particularly DeepSeek’s R1 from China earlier this year.
In its blog post on Tuesday, OpenAI said,
“We’re excited to provide these best-in-class open models to empower everyone — from individual developers to large enterprises to governments — to run and customize AI on their own infrastructure.”
Apache 2.0 Licensed and Hugging Face Hosting
Both models are licensed under Apache 2.0, enabling developers to freely download, modify, and deploy them without legal complications. The weights are already available on Hugging Face, making it easy for users to run the models locally or on cloud platforms.
Microsoft, a close partner of OpenAI, will also offer a GPU-optimized version of the gpt-oss-20B model for use on Windows devices.
Model Capabilities
- gpt-oss-120B has 117 billion parameters, with 5.1 billion activated per token.
- gpt-oss-20B has 21 billion parameters, activating 3.6 billion per token.
The number of parameters roughly correlates with a model’s intelligence and its ability to handle complex tasks. Higher parameter models generally yield better performance and understanding.
This launch is expected to significantly democratize AI capabilities, allowing users across sectors to fine-tune powerful models for domain-specific applications while maintaining full control over infrastructure and data.
With this move, OpenAI joins other AI labs in embracing open development, signaling a more collaborative and decentralized future for artificial intelligence.
