OpenAI Releases Open-Weight AI Models GPT OSS 120b & GOT OSS 20b for Public Use

OpenAI has released two open-weight AI reasoning models, GPT-OSS 120b and GPT-OSS 20b, for public use. Unlike fully open-source models, open-weight models provide trained weights but not source code or training data. Developers can access the models by downloading them from Hugging Face and deploying them locally or through Microsoft's Azure cloud platform. The shift in OpenAI's strategy comes amid growing pressure and competition in the AI space, with China's DeepSeek and Meta's Llama embracing open-source approaches.

OpenAI has released two new open-weight AI reasoning models, GPT-OSS 120b and GPT-OSS 20b, for public use. Unlike fully open-source models, open-weight models provide trained weights but not source code or training data. This shift in strategy is a response to growing pressure and competition in the AI space, with China's DeepSeek and Meta's Llama embracing open-source approaches.

GPT-OSS 120b and GPT-OSS 20b are designed to deliver strong real-world performance at low cost. Both models are available under the flexible Apache 2.0 license and outperform similarly sized open models on reasoning tasks. They are optimized for efficient deployment on consumer hardware and can run on edge devices with minimal memory requirements.

The GPT-OSS 120b model achieves near-parity with OpenAI o4-mini on core reasoning benchmarks and can run efficiently on a single 80 GB GPU. GPT-OSS 20b, on the other hand, delivers similar results to OpenAI o3-mini on common benchmarks and can run on edge devices with just 16 GB of memory. Both models also perform strongly on tool use, few-shot function calling, CoT reasoning, and HealthBench, even outperforming proprietary models like OpenAI o1 and GPT-4o.

OpenAI has trained these models using a mix of reinforcement learning and techniques informed by their most advanced internal models, including o3 and other frontier systems. The models were trained on a mostly English, text-only dataset with a focus on STEM, coding, and general knowledge. They support full chain-of-thought (CoT) reasoning and provide structured outputs.

Safety is a foundational aspect of OpenAI's approach to releasing these models. They have undergone comprehensive safety training and evaluations, including an additional layer of evaluation by testing an adversarially fine-tuned version of GPT-OSS 120b under their Preparedness Framework. The models perform comparably to their frontier models on internal safety benchmarks.

Developers can access these models by downloading them from Hugging Face and deploying them locally or through Microsoft's Azure cloud platform. The models are compatible with OpenAI's Responses API and are designed to be used within agentic workflows with exceptional instruction following, tool use, and reasoning capabilities.

OpenAI has also been working with early partners like AI Sweden, Orange, and Snowflake to learn about real-world applications of their open models. They are excited to provide these best-in-class open models to empower everyone, from individual developers to large enterprises to governments, to run and customize AI on their own infrastructure.

References:
[1] https://openai.com/index/introducing-gpt-oss/