What Is an Open-Weight Model? (Open Source AI Explained)

Updated July 23, 2026 · Data as of 2026-07-09

An open-weight model is one whose trained parameters are published for download, so anyone can run it on their own hardware or through a provider of their choosing, rather than only through the original vendor's service.

The distinction from open source is worth keeping straight, because the terms are used loosely and the difference is substantive. Open weights means you get the finished parameters. It rarely includes the training data, the training code, or the recipe, so you cannot reproduce or audit how the model came to be. Licenses vary too: some are genuinely permissive, others restrict commercial use above a size threshold or forbid particular applications. Read the license rather than the headline.

What open weights genuinely buys you is control. You can run the model on hardware you control, which matters for regulated data that cannot leave your network. Nobody can deprecate it out from under you, so a workflow you build stays reproducible for years. You can host it with whichever provider is cheapest, and you can adapt it to your own task without asking permission.

What it does not buy you is a free lunch. Running a large model well requires serious hardware and real operational work: capacity, updates, monitoring, and someone who owns it when it breaks at an inconvenient hour. For most individuals and small teams, paying a hosted provider is cheaper than the electricity plus attention, and the smaller open models that do run comfortably on a laptop are noticeably weaker than the frontier services.

Why it matters when you choose a tool

Open weights matter to you if data residency is a hard requirement, if you need a model that cannot be changed or retired beneath a workflow you depend on, or if you have enough steady volume that self-hosting beats per-token pricing. Outside those three cases it is mostly a philosophical preference, and paying for a hosted frontier model gets better results for less total effort. A middle path exists and is often the right one: use an open-weight model through a hosted provider, keeping portability without owning the hardware.

Frequently asked questions

Is an open-weight model free?

The weights are free to download, subject to the license. Running them is not free: you pay in hardware and operational time, or in fees to whichever provider hosts them for you.

Are open models as good as the frontier ones?

The best open models are strong and the gap has narrowed considerably, but the frontier hosted models generally remain ahead on the hardest reasoning and coding work. For many everyday tasks the difference is small enough not to matter.

Can I run one on my laptop?

Smaller models, yes, and quite comfortably on a modern machine with enough memory. The large ones need dedicated hardware, and a heavily compressed version that fits your laptop is meaningfully weaker than the full model.

Back to the glossary · Take the quiz · How we score tools