OpenAI, the largest generative AI company, has launched a new family of gpt-oss open-weight models offering subpar performance at a lower cost. The models are named gpt-oss-120B and gpt-oss-20B, optimized for higher efficiency, reasoning, and usability around the development environments. The new models are freely available and are here to make AI accessible for everyone.
The gpt-oss models are freely available, allowing developers to download, run, and modify them for commercial and research purposes under the open-source Apache 2.0 license. The gpt-oss-120B model is a strong competitor to OpenAI’s o4-mini; it operates on a single 80GB GPU, making it ideal for enterprise-wide applications. At the same time, the smaller gpt-oss-20B model competes with o-3 mini and can run on an edge device with 16GB of memory.
Let’s now take a look at the architecture.
GPT-OSS Architecture
Both models use a Transformer architecture and a Mixture-of-Experts (MoE) architecture. It only activates a subset of parameters per token (5.1B for the 120B model and 3.6B for the 20B model), with a top increase in throughput at inference time.
In addition to being optimized for chain-of-thought (CoT) reasoning, function calling, and tool usage that includes web surfing and Python code execution, wherein they can enable lengthy context lengths (up to 128k tokens).
OpenAI Following a Security-First Approach
OpenAI prioritizes security in its release strategy, particularly for open models. According to the tech giant, by taking the models through in-depth security training and evaluations, they have also introduced an extra layer of assessment by putting their Preparedness Framework (opens in a new window) to the test with an adversarially modified version of gpt-oss-120b.
Availability
The GPT-OSS models are available on various platforms, including Hugging Face, AWS, Azure, Vercel, and LM Studio, among others. Another notable development is that Microsoft is introducing a GPU-optimized GPT OSS-20B for Windows through the ONNX Runtime, making AI development more straightforward for Windows users.
OpenAI views gpt-oss-120b and gpt-oss-20b as significant milestones in elevating open-weight models to new heights. These models excel in reasoning and safety. This way, developers can build faster, innovate, and make a strategic move towards AI development with a range of use cases.
If you’re a developer and looking for greater control, gpt-oss has got you covered with full customization features, compatibility with OpenAI’s Responses AI, and more.
Alongside this release, it is an excellent benefit for startups, researchers, and smaller organizations that lack the financial resources to invest in these advanced models. Around the world, people may develop, invent, and create new opportunities for themselves and others by holding strong, easily accessible tools.
OpenAI's commitment is to empower the global development community and ensure the benefits of AI reach a wider audience.
Check out our news section to know all the daily tech updates right to your screen.
Also Read: A Timeline of OpenAI’s Technology and History: How OpenAI Changed the World
