GPT OSS 120B (Databricks)

GPT OSS 120B is the larger of two open-weight reasoning models released by OpenAI, issued under the permissive Apache 2.0 license that permits modification, fine-tuning, and commercial use. It is trained using a blend of reinforcement learning and techniques informed by OpenAI's most advanced internal systems, including o3 and other frontier models, which gives it a reasoning profile close to OpenAI's o4-mini on core benchmarks while still running on a single 80 GB GPU. The release also marks OpenAI's first open-source language model launch since Whisper and CLIP, signaling a deliberate return to openly distributable research artifacts.

Under the hood, GPT OSS 120B uses a mixture-of-experts design with 128 experts activated by a routing system, balancing specialization with efficient inference. It is engineered for agentic pipelines, with strong performance on chain-of-thought reasoning, tool use, and few-shot function calling as measured by evaluations such as Tau-Bench and HealthBench, where it can rival proprietary models. Configurable reasoning effort, fine-tuning support, and a very large context window make it well suited for high-volume production deployments, custom domain adaptation, and workflows that need transparent reasoning traces alongside accurate outputs.