GPT OSS 120B

GPT OSS 120B is a 117B-parameter open-weight Mixture-of-Experts language model from OpenAI, positioned for high-reasoning, agentic, and general-purpose production workloads. The MoE design lets the model scale its total parameter count while activating only a subset of experts per token, which typically yields a better quality-to-compute trade-off than dense architectures at a comparable size, making it attractive for deployments that need strong reasoning without paying for full dense inference on every call.

The model is built around configurable reasoning depth with full chain-of-thought access, allowing developers to dial how much deliberative compute the model spends on a given prompt. It also ships with native tool use that covers function calling, web browsing, and structured output generation, so it can be wired directly into agent pipelines that need to call external services, retrieve information, or emit schema-validated JSON. That combination of open weights, steerable chain-of-thought, and built-in agentic tooling makes GPT OSS 120B a practical fit for teams who want OpenAI-style reasoning behavior with the freedom to self-host, fine-tune, or inspect the model on their own infrastructure.