OpenAI has introduced the GPT-OSS, the company’s first launch of open-weight models since the release of GPT-2 in 2019.
The first gpt-oss-120b is a Mixture-of-Expert (MoE) model that activates only 5.1 billion parameters per task. The second model, gpt-oss-20b, has 21 billion parameters and activates 3.6 billion.
Both models are designed for advanced reasoning, chain of thought processing, and tool use. They serve as a low-cost alternative that is easy for researchers, developers, or businesses to implement and adopt.
Sam Altman said in an X post, “Gpt-oss is out! We made an open model that performs at the level of o4-mini. Super proud of the team; big triumph of technology.”
OpenAI described the models as “open-weight”, indicating they are not “open-source” and lack the comprehensive access to all underlying code and data.
Both models are text-only and do not support voice, images, or video inputs.
OpenAI has partnered with companies like NVIDIA, Databricks, and Microsoft to optimise its models for various hardware.
Jensen Huang, the CEO of Nvidia, gave his remarks in a blog post, “OpenAI showed the world what could be built on NVIDIA AI, and now they’re advancing innovation in open-source software.”
Extensive safety testing is conducted on the models to secure them from chemical, biological, or cyber risks. Even fine-tuned models will not be able to reach the “high capability” threshold. Sam Altman stated in a post on X, “Run additional safety tests and review high-risk areas."
The models are available on platforms like Ollama, Hugging Face, LM Studio, and within cloud providers like Bedrock, Azure AI Foundry, and Amazon.
ChatGPT is developed and owned by OpenAI, but Microsoft is a major investor that holds a significant stake.