How to Access GPT-OSS-120B: Complete Guide for Developers

GPT-OSS-120B is OpenAI’s open-weight reasoning model, released on August 5, 2025 under the Apache 2.0 licence. You can access it in three official ways: try it in the browser on the gpt-oss.com playground, download the weights from Hugging Face, or run it on your own hardware with tools like Ollama or vLLM. Running it locally needs a GPU with about 80 GB of memory.

What Is GPT-OSS-120B?

GPT-OSS-120B is the larger of OpenAI’s two open-weight models. The smaller one, GPT-OSS-20B, is built for laptops and consumer hardware. Because the weights are public, developers can run, fine-tune and deploy the model without sending data to OpenAI’s servers.

Key Specifications

• Total parameters: 117 billion, with 5.1 billion active per token.
• Architecture: Mixture-of-experts with 36 layers and 128 experts, 4 of which are active per token.
• Context length: Up to 128,000 tokens.
• Quantisation: Weights are natively stored in MXFP4, so the model fits on a single 80 GB GPU.
• Licence: Apache 2.0, which allows commercial use.

Performance and Reasoning Control

OpenAI says GPT-OSS-120B nearly matches its o4-mini model on core reasoning benchmarks and does as well or better on tests such as Codeforces, MMLU and AIME. You can set the reasoning effort to low, medium or high, trading speed for deeper answers. The model also supports tool use, such as web browsing and code execution, through the harmony prompt format.

How to Access GPT-OSS-120B

1. Try It on the gpt-oss.com Playground

The easiest option is the official gpt-oss.com playground, built by OpenAI with Hugging Face. Open the site, sign in with a Hugging Face account, choose GPT-OSS-120B and start chatting. You can switch reasoning levels to see how answers change. No setup or GPU is required.

2. Download the Weights From Hugging Face

The official model page is openai/gpt-oss-120b on Hugging Face. You can download the weights with the Hugging Face CLI or load them with the Transformers library. Make sure you use the harmony response format, otherwise the model may not respond correctly.

3. Run It Locally With Ollama

Ollama is one of the simplest ways to run the model on your own machine. Install Ollama from its official website, then run ollama run gpt-oss:120b in a terminal. The first run downloads the model, which is a large file, so use a fast and stable connection. Our guide on how to run AI models locally with Ollama explains the setup in more detail.

4. Serve It With vLLM

For production apps, vLLM can host GPT-OSS-120B as an OpenAI-compatible API server. This lets your existing apps connect to it with normal HTTP requests while your data stays on your own servers.

5. Use It Through Cloud Platforms

If you do not have an 80 GB GPU, you can use GPT-OSS-120B through OpenAI’s launch partners, such as Microsoft Azure, AWS and Databricks. These are paid services, so check each provider’s pricing before you start.

Hardware Requirements

• GPT-OSS-120B: about 80 GB of GPU memory, such as a single NVIDIA H100 or AMD MI300X.
• GPT-OSS-20B: about 16 GB of memory, which suits many gaming GPUs and Apple Silicon Macs.

If your computer cannot meet these needs, the gpt-oss.com playground is the best way to test the model.

Advantages and Limitations

Advantages: the model is open-weight, can be used commercially, keeps data private when run locally and offers strong reasoning for its size.

Limitations: it needs expensive hardware for local use, it is text-only, and like any AI model it can make mistakes, so important answers should be checked. If you want OpenAI’s latest hosted models instead, read about the GPT-5.4 launch.

Frequently Asked Questions

Is GPT-OSS-120B free to use?

The model weights are free to download under the Apache 2.0 licence, and you can test it on the gpt-oss.com playground. Running it on your own hardware or through cloud providers may still cost money.

Can I run GPT-OSS-120B on a normal laptop?

No. It needs around 80 GB of GPU memory. For laptops, OpenAI’s smaller GPT-OSS-20B model is a better choice.

Can GPT-OSS-120B be used for commercial projects?

Yes. The Apache 2.0 licence allows commercial use, modification and distribution, as long as you follow the licence terms and OpenAI’s usage policy.

Final Takeaway

GPT-OSS-120B gives developers a powerful open-weight reasoning model. Start with the gpt-oss.com playground, then move to Hugging Face, Ollama or vLLM when you are ready to run it yourself.