GPT OSS 120B (Runware)

GPT OSS 120B is an open-weight large language model created by OpenAI and built around a Mixture-of-Experts architecture, with 117 billion total parameters of which 5.1 billion activate per forward pass. Native MXFP4 quantization lets the model run efficiently on a single high-end GPU, and the weights are released under the Apache 2.0 license alongside a smaller 20B sibling, giving developers direct access to the model for self-hosting and customization. The model was trained using reinforcement learning combined with techniques drawn from OpenAI's frontier research lineage, positioning it as a flexible foundation for high-reasoning and general-purpose production workloads.

The model 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. On core reasoning benchmarks it achieves near-parity with OpenAI's o4-mini, and on agentic evaluations such as Tau-Bench and HealthBench it posts strong results where it can outperform some proprietary models. This combination of open-weight availability, efficient single-GPU inference, and agentic capability makes GPT OSS 120B well suited for developers building tool-using assistants, multi-step reasoning pipelines, and production applications that need transparency into the model's weights and reasoning process.