OpenAI Releases First Open-Weight AI Models Since GPT-2

OpenAI has released two open-weight language models, gpt-oss-120b and gpt-oss-20b, marking the company's first such release since GPT-2 in 2019. Unlike fully open-source models, these open-weight models provide access to the trained parameters (weights) but not the underlying code or training data, allowing users to run, fine-tune, and customize the models locally—including on consumer hardware for the smaller model. This move, facilitated by a permissive Apache 2.0 license, aims to meet demand from developers, researchers, and enterprises seeking greater transparency, control, and data security, particularly in response to competition from rivals like Meta, Mistral, and China’s DeepSeek. The models employ chain-of-thought reasoning, can perform tasks such as writing code and browsing the web, and are positioned as complements to OpenAI’s paid proprietary offerings. Industry experts note that while this increases accessibility and supports research, OpenAI is still not releasing fully open-source models, keeping its most advanced architectures and training methods proprietary. The launch is seen as both an effort to reassert OpenAI’s influence in the AI research ecosystem and a response to the proliferation of global open-weight alternatives.