GPT OSS 120B (SCX.ai (Turbo))

GPT OSS 120B is the larger of OpenAI's two open-weight entries in the gpt-oss family, released on August 5, 2025 under the Apache 2.0 license alongside the smaller gpt-oss-20b variant. The model is described as reaching near-parity with OpenAI o4-mini on core reasoning benchmarks, positioning it as a reasoning-focused language model that also demonstrates strong tool use, few-shot function calling, and chain-of-thought behavior as measured by evaluations such as the Tau-Bench agentic suite and HealthBench. Its training drew on a mix of reinforcement learning and techniques informed by OpenAI's own internal frontier systems, including o3.

With 120 billion parameters and a design that allows it to run efficiently on a single 80 GB GPU, GPT OSS 120B is aimed at developers who want a capable open-weight model for local or self-hosted deployments, agentic workflows, and tool-augmented applications. The Responses API compatibility highlighted in OpenAI's release makes it straightforward to integrate into agent-style pipelines, while the open weights and Apache 2.0 license allow teams to customize and fine-tune for their own domains. Through the SCX.ai Turbo route on LLM Gateway, the same underlying model is served with FP8 quantization and a long context window, giving integrators a way to access gpt-oss-120b without managing their own inference infrastructure.