The GPT-OSS 120B is a 117-billion parameter Mixture-of-Experts model engineered to deliver high-level reasoning and agentic capabilities in production environments. By activating only 5.1 billion parameters per forward pass, the architecture is specifically optimized to run efficiently on a single 80GB GPU, such as the NVIDIA H100 or AMD MI300X. It is designed to support complex developer needs, offering configurable reasoning depth and full access to the model's chain-of-thought process, which aids in debugging and transparency for specialized tasks like coding, medical analysis, and structured function calling.

Developed using a combination of reinforcement learning and techniques derived from frontier systems, this model is built to align with the harmony response format. Its design lineage emphasizes practical utility, allowing developers to leverage its reasoning strengths for agentic workflows while maintaining the flexibility of an Apache 2.0 license. By achieving performance parity with advanced proprietary models on core benchmarks, it provides a robust, cost-effective alternative for teams looking to integrate sophisticated reasoning into local or self-hosted infrastructure without the constraints of closed-source systems.