OpenAI Just Released Its First Open-Weight Models Since GPT-2
1 min read
Summary
OpenAI has released two new open-weight language models, GPT-OSS-120B and GPT-OSS-20B, which run on local devices and can be fine-tuned for specific tasks, with the release stating their aim to make AI accessible to more people.
Anyone can download these models for free from Hugging Face, a hosting platform for AI tools.
Despite being open-weight models, their ‘weights’ are publicly available for users to see, these models also differ from ChatGPT in that they don’t need to be connected to the internet to run, and can browse the web, execute code, and navigate software as an AI agent.
The smaller GPT-OSS-20B model is compact enough to run on a consumer device with more than 16 GB of memory.
Both models operate under the Apache 2.0 license, meaning they can be used for commercial purposes and redistributed as part of other licensed software.
This release reflects OpenAI’s strategy shift away from proprietary releases towards wider, more open models, and complement its paid services.
