For a lot of apps the honest answer is "not easily". So let's look at what the figures say and what you can do about it.

What Vercel's figures show

Vercel runs an AI Gateway, a service that sits between developers' apps and the various AI models they call. Every month it publishes an index of what's flowing through it. The October 2026 edition, published on 15 October and covering September, had a few numbers that stood out.

  • Open-weight models handled 71.5% of tokens. These are models whose weights are published, so anyone can run them, rather than only being available from the company that made them.
  • They're taking more of the money too. Their share of spending went from about one dollar in twenty in June to nearly one in four in September.
  • Prices are falling. The average price per token was 47% lower than in December 2025.
  • Switching hit a record. In Vercel's words, "Almost one in three teams changed their most-used lab in September". Nearly one in two changed their most-used model.

A couple of caveats, because they matter. This is only traffic that goes through Vercel's gateway, so it isn't the whole market, as The Register points out. Spending is estimated at published list prices, so real bills may differ. And the switching figures only count paid teams using at least 10 million tokens in both months, which is a fair bit of usage. A small firm's chatbot isn't in that group.

Still, the direction is clear. The AI market is moving quickly, prices are shifting month to month, and the teams who can move are moving.

Why that matters to a small business

When someone builds an AI feature, the quickest way is to pick one provider, use their software library, write the prompts to suit their model and get it working. That's fine for a demo. The trouble is that the provider ends up woven through the whole codebase.

Then something changes. A cheaper model comes out that does the job just as well. Your provider raises prices, retires the model you were using or changes how it behaves. Or a customer asks where their data is processed and you'd rather it stayed somewhere specific. If swapping means rewriting half the app, you'll probably just put up with it and keep paying.

That's the real lesson from those switching figures. It isn't that everyone should be chasing the latest model every month. It's that the teams who switched were able to. That ability is worth having, even if you never use it.

How to keep your options open

None of this is complicated or expensive if it's done early. Most of it is just tidy engineering.

  • Keep the AI calls in one place. Your app should talk to its own small piece of code that handles AI, and only that piece should know which provider you're using. Then a switch is a change to one module, not a hunt through the whole system.
  • Make the model a setting. Which model and which provider should live in configuration, not be typed into the code in fifteen places.
  • Write down what "good" looks like. Keep a set of real examples with the answers you expect. When you try a different model, run it against those and compare. Without this, switching is guesswork, and guesswork is how you end up with a cheaper model that quietly gets things wrong.
  • Don't lean on one provider's special features unless you need them. Some are worth it. Just make that a decision, not an accident.
  • Log usage and cost per feature. You can't tell whether a switch saved money if you never knew what you were spending. We wrote about that in more detail in how to stop your AI bill running away.
  • Know where your data goes. Open-weight models can be run by lots of different hosting companies, or on your own servers. That's useful, but check who's running yours and where before you send customer data to it.

Is open-weight right for you?

Maybe, maybe not. Those Vercel figures show open-weight models doing most of the volume but well under half the spending, which tells you they're being used heavily for high-volume work where price matters. Plenty of teams still pay more for the big commercial models on the jobs where quality matters most.

That's the sensible approach for most businesses too. Use the cheaper model where it does the job and the better one where it earns its keep. You can only do that if your app is built to let you choose.

Where we come in

When we build software with AI features, the client owns the code outright and the architecture decisions are made by people, not the model. Keeping the AI provider swappable is part of that.

If you've already got an AI app, perhaps built quickly with one of the AI tools, our AI Code Audit will tell you how it's put together, including how tied it is to one provider. You get a written report in plain English within five working days, with findings rated by severity and a prioritised plan for fixing them.

And if you'd just like someone keeping an eye on all this, along with your other suppliers and subscriptions, that's what our fractional CTO service is for.

The takeaway

The AI market is shifting fast, and almost a third of busy teams changed their main provider in a single month. You don't need to follow them. You just need to be able to. If you'd like a straight answer on how locked in your app is, give us a ring on 01474 639 089 or drop us a line.

Cheers, Rob