GPT-OSS 120B
GPT-OSS 120B is OpenAI's open-weight, text-only reasoning model for coding, knowledge work, and tool-driven agents. Its Mixture-of-Experts architecture contains 117B total parameters but activates only 5.1B per token, with a 128K-token context window for long documents and complex tasks. The model supports multi-step reasoning, function calling, structured outputs, and workflows that connect to external tools such as search or Python execution. Reasoning effort is configurable at low, medium, or high to balance depth and latency. Alternating dense and sliding-window attention supports long-context processing, while MXFP4-quantized expert weights allow the model to fit on a single 80GB GPU. It uses OpenAI's Harmony conversation format, and is released under Apache 2.0 for commercial use.
GPT-OSS 120B
GPT-OSS 120B is OpenAI's open-weight, text-only reasoning model for coding, knowledge work, and tool-driven agents. Its Mixture-of-Experts architecture contains 117B total parameters but activates only 5.1B per token, with a 128K-token context window for long documents and complex tasks. The model supports multi-step reasoning, function calling, structured outputs, and workflows that connect to external tools such as search or Python execution. Reasoning effort is configurable at low, medium, or high to balance depth and latency. Alternating dense and sliding-window attention supports long-context processing, while MXFP4-quantized expert weights allow the model to fit on a single 80GB GPU. It uses OpenAI's Harmony conversation format, and is released under Apache 2.0 for commercial use.
