GPT OSS 120B is OpenAI's open-weight large language model built around a Mixture-of-Experts design, with 117 billion total parameters of which 5.1 billion activate per forward pass, allowing it to run efficiently on a single high-end GPU using native MXFP4 quantization. It is distributed under the Apache 2.0 license alongside its smaller sibling, the 20B variant, giving developers full access to the weights. The model was trained with a combination of reinforcement learning and techniques drawn from OpenAI's frontier research lineage, including o3, and is positioned as a flexible foundation for high-reasoning, agentic, and general-purpose production workloads.

In practice, GPT OSS 120B is engineered for agentic workflows, offering configurable reasoning depth, full chain-of-thought access, and native tool use including function calling, browsing, and structured output generation. It achieves near-parity with OpenAI's o4-mini on core reasoning benchmarks and posts strong results on agentic evaluations such as Tau-Bench and HealthBench, where it can outperform some proprietary models. The Responses API compatibility and broad ecosystem support make it well suited for teams building multi-step agents, code-execution pipelines, or research applications that need transparent reasoning traces without the constraints of a closed-source deployment.