Open-weight model

What is an open-weight model?

An open-weight model is an AI model whose trained weights are published, so companies can download, run and adapt it on their own infrastructure. Weights are the numerical parameters that encode what a model learned in training. Publishing the weights does not always mean publishing the training data or code, and licenses can limit some uses.

Why it matters to a revenue leader

For most revenue teams, the choice of model sits with the software vendor. It becomes the revenue leader's question when customer data must stay inside the company's own environment, or when inference cost at high volume makes running your own model attractive. Running an open-weight model moves work to your team: hosting, security, updates and testing. In July 2024 the U.S. Commerce Department's NTIA concluded that current evidence was insufficient to warrant restricting these models and recommended that government actively monitor their risks.

An environmental services platform owned by a PE sponsor holds sensitive site and contract data under strict customer agreements. Its IT team runs an open-weight model inside the company's own cloud account to summarize service agreements, so no contract text goes to an outside model provider. The trade-off is that the team now owns updates, security patches and quality testing for that model.