THE INSTITUTE / RESEARCH TOPICS
Open-weight models
What you can download, modify, and run yourself—and what the license allows.
Explore this topic
Open-weight models
Choose a layer to understand the tools and buying questions.
The quick answer
An open-weight model makes its learned parameters available to download. With compatible software and enough computing resources, a team can run a released model itself. The license determines what uses, modifications, and redistribution are allowed.
Downloadable weights do not automatically mean the entire training process is available, unrestricted commercial use is allowed, or the model will be inexpensive to operate. Check the specific release and license.
What it is
Weights are the learned numerical parameters used by a model. Releasing them can give a team more control over where the model runs and how it is adapted. It also shifts responsibilities: the team or its chosen provider must handle serving, updates, security, and capacity.
Know the difference
Open weights and open source are not identical
The Open Source Initiative’s AI definition considers more than the availability of weights, including code, data information, and rights to use and modify the system. Treat “open” as a starting point for questions, not a complete specification.
Self-hosting and hosted access can coexist
A downloadable model can also be served by a cloud provider. Decide separately which model to use and who will operate it.
How it works
Check the release
Identify the model version, available weights, supporting software, and license.
Choose the operating model
Run it on infrastructure your team manages, or buy a hosted service. Confirm responsibilities and where data is processed.
Test the actual deployment
Model size, numerical precision, hardware, and serving settings can change the deployment’s behavior and economics. Evaluate the setup you intend to use.
Types of tools
Platforms and tools
Business uses
Benefits and costs
Potential benefits include deployment choice and control over the serving environment. Costs include computing capacity, engineering, monitoring, maintenance, and any commercial support. A free download is not a zero-cost service.
Limits and risks
What the evidence shows
How we assess it
This guide combines primary documentation with the Institute’s questions for a buying team. It does not report a survey, independent benchmark, or vendor score.
Questions before you buy
- What exactly has been released, and under which license?
- Does the license permit our planned use and distribution?
- What hardware and operating skills are required?
- Who applies updates and investigates failures?
- Can we reproduce quality and latency on our intended deployment?
- How does total cost compare with a hosted alternative?
Common questions
Sources and reading
Open Source Initiative — Open Source AI Definition 1.0: one explicit standard for open-source AI.
OpenAI — gpt-oss: weights, license, and access-route example.
Google — Gemma overview: a downloadable model family.
IBM — Granite 4.2: an Apache 2.0 model release.
This guide explains buying considerations. It does not replace the license or contract for a particular release.
What changed
October 24, 2026 — First publication as part of the Models research category.