← Back to Feed

Post from 19 October 2026

Adding an AI feature to a product usually starts with picking a vendor’s API.

That one choice decides two things: which model you use, and where it runs. With a closed model like GPT or Claude, the vendor or one of its cloud partners runs it. You can often pick a region, but your data still leaves your own systems.

Open-weight models separate those decisions. The trained model is published, so the same model can run in several places. A cloud provider can host it for you, you can run it on your own GPUs. You can also further train a copy on your own examples (fine-tuning). Each option trades off cost, speed, security, who controls the data, quality and upkeep differently.

That matters most for a feature over data whose location is regulated. If the rules say the data can’t leave your systems, that can rule out options before you ever compare benchmarks.

This assumes an open model is good enough for the task and your team can operate one. Self-hosting isn’t automatically better.

So before you compare models, find out where your data is allowed to go. The answer may rule out some models and hosting options before you start.

Also on Bluesky · Mastodon

Loading comments…