OpenAI has released two new open-weight language models, gpt-oss-120b and gpt-oss-20b, on August 5, 2025, signaling a significant strategic shift towards open-source AI development. These models, available under the Apache 2.0 license, represent OpenAI's first open-weight releases since GPT-2, allowing developers and enterprises to run and adapt them on their own hardware. The announcement has been met with enthusiasm from the AI community, including a positive social media post from Dominic Phillips.
The gpt-oss-120b model boasts 117 billion total parameters with 5.1 billion active parameters per token, designed for high-reasoning tasks and capable of running efficiently on a single 80GB GPU like an NVIDIA H100. Its smaller counterpart, gpt-oss-20b, features 21 billion total parameters with 3.6 billion active parameters, optimized for lower latency and local or specialized use cases, even running on consumer hardware with 16GB of memory. Both models leverage a Mixture-of-Experts (MoE) architecture and MXFP4 quantization to enhance efficiency.
According to OpenAI, the gpt-oss-120b model achieves performance near parity with its proprietary o4-mini model on core reasoning benchmarks, including competition mathematics and agentic evaluations. The gpt-oss-20b model is comparable to o3-mini, and in some instances, surpasses its performance. This performance-to-size ratio makes them highly versatile for various deployment scenarios, from cloud-based enterprise solutions to on-device AI applications.
The models are readily accessible on platforms such as Hugging Face, Azure AI Model Catalog, and Ollama, facilitating widespread adoption and experimentation. Dominic Phillips, in a tweet, expressed satisfaction with the new models, stating, "Came preloaded with gpt-oss 20B & 120B. Nice touch. Thanks for the fun event! @OpenAI @edwinarbus." This highlights the immediate usability and positive user experience of these newly released models.
This release underscores OpenAI's renewed commitment to democratizing AI by providing powerful, adaptable models that can be deployed privately and securely without reliance on cloud-based APIs. The open-weight nature of GPT-OSS 120B and 20B is expected to foster innovation within the AI ecosystem, enabling a broader range of developers to build and fine-tune advanced AI applications.