GPT OSS 120B is OpenAI's large-scale open-weight language model, released in August 2025 alongside its smaller sibling. It was trained using techniques informed by OpenAI's most advanced internal systems, including the o3 reasoning model, with reinforcement learning applied to strengthen chain-of-thought capabilities and tool use. The model uses a Mixture-of-Experts architecture with 128 active experts routed dynamically, letting it deliver reasoning performance comparable to OpenAI's o4-mini on core benchmarks while fitting on a single 80 GB GPU for data-center deployments. Distributed under the Apache 2.0 license, it can be run locally, modified, fine-tuned, and commercialized, making it a practical option for teams that need frontier-style reasoning without proprietary lock-in.
Beyond raw reasoning, GPT OSS 120B is designed to slot into agentic workflows, supporting configurable reasoning effort levels (low, medium, high) and strong few-shot function calling. In OpenAI's evaluations it performed well on the Tau-Bench agentic suite and HealthBench, even surpassing proprietary models like o1 and GPT-4o on those assessments, which signals solid fit for production tool-calling and multi-step task automation. Through Eden AI, developers can access the model alongside 16 other providers tracked on Artificial Analysis, where Cerebras leads on combined speed and latency while CoreWeave is positioned as the most cost-competitive option, giving teams meaningful flexibility to balance throughput, latency, and price for their workloads.
