OpenAI has unveiled two new generative AI models: gpt-oss-120b and gpt-oss-20b. While not the highly anticipated GPT-5, these open-weight models are a notable advancement, marking OpenAI’s first open-weight release since GPT-2 in 2019. Both models are designed for text-only tasks, leveraging chain-of-thought reasoning to enhance problem-solving by breaking complex tasks into manageable pieces. They can browse the web and work with cloud-based solutions. The gpt-oss-20b model, notable for its compatibility with PCs powered by Snapdragon processors, enables on-device inference—an industry first. Supported by NVIDIA H100 GPUs, the models required extensive training, with gpt-oss-120b taking 2.1 million hours. Users can download these models through Hugging Face, and NVIDIA RTX GPUs can provide optimal performance. This release strengthens OpenAI’s position in the AI landscape, encouraging broader access and innovation.
Source link
Share
Read more