OpenAI has released two state-of-the-art open-weight language models, gpt-oss-120b and gpt-oss-20b, which push the boundaries of open-weight reasoning models. These models deliver strong real-world performance at low cost and are available under the flexible Apache 2.0 license.
gpt-oss-120b achieves near-parity with OpenAI o4-mini on core reasoning benchmarks while running efficiently on a single 80 GB GPU. gpt-oss-20b delivers similar results to OpenAI o3-mini on common benchmarks and can run on edge devices with just 16 GB of memory, making it ideal for on-device use cases.
The models have been post-trained using a combination of reinforcement learning and techniques informed by OpenAI’s most advanced internal models. They demonstrate exceptional capabilities in tool use, few-shot function calling, CoT reasoning, and HealthBench, even outperforming proprietary models like OpenAI o1 and GPT-4o.
Safety is foundational to the approach of releasing these models, with comprehensive safety training and evaluations conducted on both models. The gpt-oss models perform comparably to OpenAI’s frontier models on internal safety benchmarks, offering developers the same safety standards as recent proprietary models.
Developers can use these models within agentic workflows, leveraging exceptional instruction following, tool use like web search or Python code execution, and reasoning capabilities, including adjusting the reasoning effort for tasks that don’t require complex reasoning. The models are entirely customizable, provide full chain-of-thought (CoT), and support Structured Outputs.
To contribute to a safer open-source ecosystem, OpenAI is hosting a Red Teaming Challenge with a $500,000 prize fund to encourage researchers, developers, and enthusiasts to identify novel safety issues. The challenge aims to foster innovation, enable safer AI development, and promote democratization of AI tools for emerging markets and resource-constrained sectors.
The gpt-oss models are freely available for download on Hugging Face, with native quantization in MXFP4, allowing them to run efficiently on various hardware configurations. This release marks a significant step forward for open-weight models, delivering meaningful advancements in both reasoning capabilities and safety while promoting accessibility and democratization of AI tools worldwide.
Source: https://openai.com/index/introducing-gpt-oss