One of the first architectural decisions in any AI project is also one of the most consequential: build on a closed model behind an API, or an open-weight model you run yourself? Here is how to think about it.
Closed API models
Models accessed through a provider’s API offer frontier capability with almost no setup — you send a request and get an answer. The provider handles hosting, scaling, and updates. The trade-offs: you pay per use, your data leaves your infrastructure (subject to the provider’s terms), and you cannot inspect or deeply customize the model.
Open-weight models
Open-weight models ship their parameters, so you can run them on your own hardware or cloud. That gives you control (customize and fine-tune freely), privacy (data never leaves your environment), and no per-token fees. The cost is operational: you own the hosting, scaling, and maintenance, which is real work.
The capability picture
The old assumption that open models are far behind no longer holds broadly. Strong open-weight models now handle a large share of real-world tasks well. For the very hardest reasoning or newest frontier features, the best closed models often still lead — but the gap has narrowed enough that “open is good enough” is true for many use cases.
How to choose
- Prototyping or need top capability fast? Start with a closed API.
- Strict privacy, high volume, or cost sensitivity? Lean open-weight.
- Both? Very common — route easy, high-volume, or sensitive work to open models and reserve closed frontier models for the hardest cases.