The term \u201copen model\u201d covers very different situations, and the differences matter exactly when you are making a decision.
What \u201copen\u201d can mean
Published weights, permissive licence. You can download the model, run it where you like, modify it and use it commercially.
Published weights, restrictive licence. You can download it, but the licence limits commercial use, application domains or scale.
Fully open. Alongside the weights, the training data, the code and the procedure are published — which makes reproduction possible. This is the rarest case.
Interface only. The model runs at the provider; you get API access. You neither see nor control anything inside.
Many models presented as \u201copen\u201d are in the second category. Check the licence, not the press release.
The criteria that matter operationally
Where the data goes. If you process sensitive data and cannot or will not send it outside your own infrastructure, a locally run model is sometimes the only option. It is the most common real reason organisations choose open models.
Stability over time. A model accessed through an API can change without your say. Behaviour shifts, and the system built on top behaves differently. A downloaded model stays identical as long as you keep it.
Cost at scale. Below a certain volume the API is almost certainly cheaper. Above a threshold, running it yourself becomes competitive — but the calculation includes operations, not only hardware.
Latency and availability. Running locally removes the dependency on the network and on an external provider's uptime.
Operational effort. A locally run model needs someone to administer it, update it and respond when it breaks at two in the morning. For small teams this is usually the deciding argument.
Capability. The performance gap between the best closed models and the best open ones varies over time and by task. For many practical tasks a smaller open model is sufficient — worth testing before assuming otherwise.
A criterion often overlooked
The ability to audit. If you have to demonstrate to an auditor or a regulator how the system works, a model you fully control puts you in a much better position than one about which you can only repeat the supplier's documentation.
The conclusion
The question is not \u201copen or closed\u201d, but: what constraints do we have on data, stability, cost and audit — and what can we realistically operate? The answer differs between organisations and can change in six months.