× News Alerts AI News CyberSec News Let's Talk Local AI Bank Tech News Cyber Advisories Contact

OpenAI Releases New Open-Weight Reasoning Models

OpenAI has released two new open-weight AI models, gpt-oss-120b and gpt-oss-20b, aimed at advancing reasoning in open-source AI. The models are designed for developers, offering powerful capabilities on accessible hardware and are available under the Apache 2.0 license, marking a significant strategic shift for the company.

OpenAI Releases New Open-Weight Reasoning Models

OpenAI has announced the release of two new, state-of-the-art open-weight language models, gpt-oss-120b and gpt-oss-20b, marking a significant shift in the company's strategy. This move comes more than five years after its last open language model release, GPT-2, and positions OpenAI in more direct competition with open-source leaders like Meta and Mistral AI.

The new models are designed to advance the frontier of reasoning in open-source AI, providing powerful tools for developers and researchers. The models are available under the flexible Apache 2.0 license, allowing for free use, modification, and commercial deployment. OpenAI CEO Sam Altman described the release as an effort to "get AI into the hands of the most people possible," emphasizing the company's commitment to democratizing AI innovation.

The two models differ in size and hardware requirements. The gpt-oss-120b, which boasts 117 billion parameters, can be run on a single Nvidia GPU, while the lighter gpt-oss-20b model is optimized for consumer laptops equipped with 16GB of memory. This accessibility makes the models suitable for practical, real-world applications. Both models use a Mixture-of-Experts (MoE) architecture and a 4-bit quantization scheme for fast inference with low resource usage.

According to OpenAI, the gpt-oss models were trained using a mix of reinforcement learning and techniques informed by OpenAI's most advanced internal models. They are designed to outperform similarly sized open models on reasoning tasks, demonstrate strong tool use capabilities, and are optimized for efficient deployment on consumer hardware. NVIDIA collaborated with OpenAI on these open models, which were trained on NVIDIA H100 GPUs.

Despite the "open-weight" designation, the models are not fully transparent. Critical elements such as the routing mechanisms and training methods remain confidential. This approach allows OpenAI to provide a useful tool to the public while protecting its valuable intellectual property. The release is seen as a meaningful step in the company's commitment to the open-source ecosystem and its mission to ensure AI benefits all of humanity.

Subscribe for AI & Cybersecurity news and insights