In a bold move to bring artificial intelligence closer to everyday users, OpenAI has released two offline AI models designed to function without cloud services or internet access. The gpt-oss-120b and gpt-oss-20b models offer powerful language generation capabilities while running directly on consumer-level hardware.

This initiative, announced by OpenAI CEO Sam Altman, signals a shift toward edge AI—an approach that decentralizes computing and enables localized intelligence.

“These are the most useful open models in existence,” Altman shared on X. “We’re proud to make them available to anyone, anywhere.”

Aimed at reducing environmental impact, the models operate with far fewer resources than conventional data center-based systems. However, the tradeoff is their text-only output, with no image or audio generation at this stage.

The models were developed with security in mind. OpenAI collaborated with third-party safety organizations to evaluate potential misuse, including threats related to cybersecurity and bioengineering.

Altman highlighted the research potential these tools unlock, saying, “We’re going to see more people than ever before doing meaningful work with AI—work that was out of reach until now.”

The release represents the company’s first open-weight model launch since GPT-2, responding to increasing demand for customizable and private AI tools.

OpenAI has made both models accessible for download on developer platforms like Hugging Face and Databricks, allowing developers and researchers to start experimenting immediately.