OpenAI Releases GPT-OSS

Here's everything you need to know about GPT-OSS model from OpenAI. Also learn the different ways to access them for free.

Aug 12, 2025

∙ Paid

OpenAI just released its first fully open-source large language model or LLM family under an Apache 2.0 license. They’re calling it GPT-OSS or Generative Pre-trained Transformer (Open Source Software).

The release includes two models:

GPT-OSS-20B: 21 billion total parameters (around 3.6 billion are active), fits in 16 GB of VRAM
GPT-OSS-120B: 117 billion total parameters (around 5.1 billion active), can run on a single H100 or multiple
Generative AI Publication is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

They share the same architecture but are optimized for different hardware levels. Both are mixture-of-experts (MoEs) and use a 4-bit quantization scheme (MXFP4), enabling fast inference while keeping resource usage low.

In terms of size, the larger model can run on a single H100 GPU, while the smaller one fits within 16GB of memory, which makes it small enough to run on consumer hardware.

Key Features of GPT-OSS Models

These new models are designed to be used within agentic workflows. They work really well with the following tasks:

Keep reading with a 7-day free trial

Subscribe to Generative AI Publication to keep reading this post and get 7 days of free access to the full post archives.