OpenAI has introduced its first open-weight language models, GPT-OSS 20B and GPT-OSS 120B, designed for local AI processing. The release includes a 116.8 billion parameter model, referred to as 120B, with 5.1 billion active parameters, and a 20.9 billion parameter model, referred to as 20B, with 3.6 billion active parameters.
Both models are supported on AMD’s Ryzen AI processors and Radeon graphics cards, allowing users to run advanced AI tasks on personal computers instead of relying on data center hardware. The models can be accessed through OpenAI’s enablement partner, LM Studio, on compatible devices.
AMD announced that the Ryzen AI Max+ 395 is the first consumer AI PC processor capable of running the GPT-OSS 120B model. The model requires about 61GB of video memory, which fits into the 96GB dedicated graphics memory of the Ryzen AI Max+ 395. Users need AMD Software: Adrenalin Edition 25.8.1 WHQL or later to activate the feature.
The processor can generate up to 30 tokens per second, with performance aided by its high memory bandwidth and the mixture-of-experts architecture used in the GPT-OSS 120B model. It also supports Model Context Protocol (MCP) for managing larger AI workloads. Owners of Ryzen AI 300 series processors can run the smaller GPT-OSS 20B model.
For faster output with the 20B model, AMD recommends pairing it with the Radeon 9070 XT 16 GB graphics card in a desktop setup. This combination improves both token generation speed and time-to-first-token, which benefits applications that require quick AI responses.
AMD said the expanded support allows more users to access powerful AI capabilities on standard consumer hardware.