Open-weight models matured fast through 2026 and are the right call when data cannot leave your infrastructure or per-request costs must approach zero. API frontier models remain the default for everything else: less operations burden, faster improvement, better peak capability.
The debate generates more heat than the decision deserves. It is a build-vs-buy question wearing a philosophy costume.
APIs win on capability ceiling, zero maintenance and pace of improvement: you wake up to a better model without lifting a finger. Open weights win on data locality, cost at scale, and independence from a vendor's pricing and policy changes. Both are legitimate, for different constraints.
A rule that survives contact
Choose open weights when a regulator, client contract or your own risk posture forbids data leaving your control, or when volume makes per-token pricing the dominant cost. Otherwise start with APIs, version-pinned, behind an abstraction layer that lets you switch later.
For a typical SMB the sequencing is almost always API first, prove the workflow, then re-evaluate hosting when the invoice or the compliance question actually arrives.