GPT OSS 120B

GPT OSS 120B is the larger of two open-weight reasoning models released by OpenAI in the gpt-oss family, paired with a smaller 20B sibling. Both models are distributed under the Apache 2.0 license, which permits local or cloud hosting, fine-tuning, modification, and commercial use, marking OpenAI's first open-source language model release since Whisper and CLIP. The 120B variant is intended for production, general-purpose workloads that demand stronger reasoning depth than the 20B version, which is targeted at edge devices and laptops with 16 GB of memory. Through Cloudflare's hosted inference, the model becomes accessible without standing up dedicated GPU infrastructure, broadening its reach for developers building on the Workers AI platform.

Training drew on a mixture of reinforcement learning and techniques informed by OpenAI's more advanced internal systems, including o3 and other frontier models, which shaped its tool use and chain-of-thought behavior. OpenAI positions GPT OSS 120B as reaching near-parity with o4-mini on core reasoning benchmarks while running efficiently on a single 80 GB GPU, and reports strong results on tool use, few-shot function calling, and agentic evaluations such as Tau-Bench, where it reportedly outperforms proprietary peers like o1 and GPT-4o on HealthBench. These characteristics make the model a practical fit for agentic workflows, instruction-heavy applications, and teams that want open-weight flexibility combined with reasoning and function-calling competence.