GPT OSS 120B (Fireworks AI)

GPT OSS 120B is OpenAI's larger entry in the open-weight gpt-oss family, designed for production reasoning and agentic workloads. It carries 117B total parameters with 5.1B active at inference, a sparse mixture-of-experts-style configuration that lets the full model fit on a single 80GB accelerator such as an NVIDIA H100 or AMD MI300X. Training drew on reinforcement learning and techniques informed by OpenAI's own frontier systems, including o3, and the model is tuned to operate within the harmony response format that OpenAI ships alongside the release.

In practice, GPT OSS 120B is positioned for developers who want OpenAI-style reasoning under permissive terms, with the Apache 2.0 license enabling commercial deployment, fine-tuning, and customization. OpenAI reports near-parity with o4-mini on core reasoning benchmarks, and the model card highlights configurable reasoning effort, full chain-of-thought visibility for debugging, and strong function-calling and tool-use behavior that support agentic pipelines. Compared with its 20B sibling, it trades edge-device efficiency for higher reasoning capacity, making it the better fit when a single high-memory GPU is available and the workload benefits from deeper deliberation.