The Core Difference
An open-weight (often loosely called "open-source") model has its trained parameters publicly downloadable, letting anyone run it on their own infrastructure, modify it, and inspect (to some extent) how it behaves. A closed model is only accessible through the provider's own API or app — you can use it, but you can't download or directly inspect its weights.
Advantages of Open-Weight Models
Open-weight models offer more control over deployment (including running fully offline or on private infrastructure), no per-token API cost once you're hosting it yourself, and the ability to fine-tune or modify the model directly for your specific use case.
Advantages of Closed Models
Closed models are typically simpler to start with (no infrastructure to manage), often represent a provider's most capable, frequently updated offering, and come with a managed API that handles scaling and reliability without you needing to run your own servers.
How to Decide
The choice usually comes down to your specific constraints: if you need full control, data privacy through self-hosting, or want to avoid ongoing per-token costs at high volume, open-weight is worth serious consideration. If you want the simplest path to a working product and don't want to manage infrastructure, a closed model's managed API is usually the faster route.
Related Pages
Frequently Asked
Is 'open-source' the technically correct term for these models?
Strictly, "open-weight" is more precise, since the trained parameters (weights) are released, but the full training process, code, and data aren't always included; "open-source" is commonly used more loosely in casual conversation.
Are open-weight models always free?
The weights themselves are typically free to download, but running them still requires computing infrastructure, which has its own real cost.
Which is more capable, open-weight or closed models?
This shifts over time as both categories improve; leading closed models often have a capability edge at any given moment, but the gap has narrowed significantly as open-weight releases have improved.
Can I fine-tune a closed model?
Some providers offer limited fine-tuning through their API for closed models, though you typically have far more control and flexibility when fine-tuning an open-weight model you're hosting yourself.