Generative AI Tools

Generative AI tooling splits along one main line: hosted APIs you call, and open-weight models you run. Hosted providers give you the strongest models with no infrastructure to manage, billed per token. Open weights cost more effort and hardware but keep your data in-house and remove per-call pricing — which matters once volume grows or the data is sensitive.

The second thing worth deciding early is how much you want the provider to own. A raw completion API leaves orchestration, retries, evaluation and prompt management to you. Higher-level platforms bundle those, at the cost of portability. Teams that expect to switch models tend to keep a thin abstraction and lose the convenience.

Pricing here moves faster than almost any other category, and per-token rates published today are frequently stale within months. Treat cost as something to re-check against the vendor's own pricing page rather than a fixed property of the tool.