What is Open-Weight AI Model?

An open-weight model is an AI model whose trained parameters—the weights—are released so that anyone can download and run it on their own hardware. This contrasts with models available only through a company’s API.

Open weights are not the same as open source

Publishing weights does not automatically mean the model is “open source” in the traditional sense:

• The training data and code may not be released.
• The license decides what you may do. Some models use permissive licenses such as Apache 2.0 or MIT, which allow commercial use. Others use research-only or custom licenses that forbid commercial use or add conditions such as user limits or attribution.

Always read the license on the model card before using a model—or its outputs—in a commercial product.

Why people use them

• Control and privacy: data can stay on your own servers.
• Cost: running a model locally can be cheaper at scale than paying per API call.
• Customization: weights can be fine-tuned for specific tasks.

Trade-offs

Large models need powerful GPUs with lots of memory, and self-hosting means handling updates, safety measures, and performance tuning yourself. Hardware such as high-bandwidth memory (HBM) determines how large a model can be run efficiently.

This glossary entry is general information.