Closed vs Open-Weight Models
Closed models are served only through a provider's API with weights never released; open-weight models publish their weights for anyone to download, run, and fine-tune.
| Dimension | Closed (API-Only) Models | Open-Weight Models |
|---|---|---|
| Time to a production system | Wins here Call an API — no infrastructure to stand up or maintain. | Requires provisioning and operating your own serving infrastructure before the first request. |
| Data control and privacy | Requests leave your infrastructure and go to a third-party provider. | Wins here Can run entirely on infrastructure you control, with data never leaving your boundary. |
| Customization depth | Typically limited to prompt-level control and, at most, a constrained fine-tuning API. | Wins here Full fine-tuning, quantization, and architecture-level changes are possible once you hold the weights. |
| Frontier capability | Wins here The most capable models at any given time are usually closed, though the gap narrows and shifts over time. | Strong, but generally trails the closed frontier on the hardest tasks at any given moment. |
| Operational burden | Wins here The provider handles scaling, serving infrastructure, and model updates. | You own GPU provisioning, quantization decisions, and uptime. |
| Cost at scale | Usage-based pricing is predictable but can dominate spend at very high, sustained volume. | High fixed infrastructure cost upfront, but can be cheaper per token at sufficient sustained scale. |
| Model continuity | A model behind a "latest" alias can change under you without warning — a real source of model drift. | Wins here A downloaded checkpoint does not change unless you choose to swap it. |
When to choose Closed (API-Only) Models
Choose a closed model when you need frontier capability without an infrastructure investment, or your scale doesn't justify self-hosting costs.
When to choose Open-Weight Models
Choose an open-weight model when data must stay on infrastructure you control, you need deep customization, or you're optimizing hard for cost-per-token at high, sustained volume.
The verdict
For most teams starting out, a closed API is the pragmatic default. Open-weight models earn their added complexity once data residency, deep customization, or sustained scale make self-hosting pay for itself.