Glossary term

Open weights

A release in which a model's trained weights are available to download.

How it works

With an open-weights release, the trained model parameters are published for anyone to download, run, inspect and often fine-tune. Training code and data are usually not included, and the license may restrict some uses.

Open weights let developers run models on their own hardware.

Using an open-weights model

Step 1Download the weights
Step 2Load into a framework
Step 3Run locally or in your cloud
Step 4Optionally fine-tune

In code

Load open weights from a hub

from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("<model-id>")
model = AutoModelForCausalLM.from_pretrained("<model-id>")

Why it matters

They shape competition and access to AI, and raise debates about safety when powerful models cannot be recalled.

Related terms

Open source Model weights Model hub