OpenAI has launched gpt-oss, a groundbreaking AI large language model (LLM) for local download and offline use, licensed under Apache 2.0. This is OpenAI’s first open release, boasting features that compete with their hosted services. Partnering with ollama, an MIT-licensed open-source project, streamlines the local LLM installation process. Two models are available: gpt-oss-20b, requiring 16 GB of memory, and the more powerful gpt-oss-120b, needing 80 GB. OpenAI claims the smaller model parallels their hosted o3-mini “reasoning” model, while the 120b variant outperforms it. Unlike other models, this release marks OpenAI’s debut of an LLM. Additionally, a bounty challenge has been introduced, offering $500,000 for identifying flaws in gpt-oss-20b, with submissions due by August 26, 2025. This combination of local capability and financial incentive makes gpt-oss a noteworthy development in the AI landscape.
Source link
Share
Read more