Open-weight models

Open-weight models are AI models whose trained weights are published, so anyone can download, run and change them. Examples include Llama and DeepSeek.

AI searchUpdated 1 min read

Also called open-source LLMs · open models · open-source AI

What it means

A model's weights are the numbers it learned in training. With closed models like GPT, only the company has them. With open-weight models, they're published.

Popular open-weight models include Meta's Llama, DeepSeek, Mistral and Alibaba's Qwen. They're often called "open source", though their licences vary.

Why it matters for brands

Open models end up inside many products: customer service bots, shopping assistants, search features and apps. When they answer a question about your category without searching, they rely entirely on what they learned in training.

That learning is frozen at the time the model was trained. If you weren't well covered online then, these models won't know you.

What you can do

You can't update a model's training. But building a clear, consistent presence on widely read sites makes it more likely future models learn about you.

Example

A shopping app uses an open-weight model to answer "which robot vacuum should I buy?" without searching the web. It names models from two years ago, because that's what its training data covered.

Frequently asked questions

Are open-weight models as good as ChatGPT?

The best are close for many tasks. The products around them, like web search, matter as much as the model.

Can I add my brand to an open model?

Companies running the model can fine-tune it or connect it to search. You can't change the public model.

Is open-weight the same as open source?

Not exactly. Many open-weight models don't publish their training data or code, and some limit how they can be used.

Go deeper

All posts →

Make AI search your next revenue channel

Start free trial