← Back to briefings

Vercel AI Gateway Leaderboards Add Open Data

2026-07-13 • July 13, 2026 • Butler

AI Gateway leaderboards matter more now that the data can be shared, downloaded, and queried, because model-routing arguments can move from vibes and screenshots toward repeatable evidence.

A butler comparing ranking boards while annotating a chart for different service routes

AI leaderboard discourse usually gets stuck in screenshot mode.

Somebody posts a chart, everybody argues about what it means, and almost nobody can do much with the underlying data.

Vercel's July 13 update to AI Gateway leaderboards is more useful than that. The company says the leaderboards now support shareable charts, downloadable CSV exports, and a queryable export endpoint, with the data published under CC BY 4.0.

That does not make the rankings universal truth. It does make them more reusable.

The real upgrade is from display data to working data

Vercel says the leaderboards rank models, labs, apps, and providers running on AI Gateway, using production traffic aggregated daily across trillions of tokens.

That is already interesting. But the better story is what happens once the data stops being trapped in the interface.

When a chart can be exported, shared, and queried, teams can do more than point at it. They can compare trends over time, pull the data into internal analysis, and stop relying on hand-cropped screenshots as if they were evidence.

The shift is small in product terms and big in conversation quality.

There are now clearer ways to ask practical routing questions

Vercel lists four leaderboard families:

It also says the metrics vary by leaderboard, including requests, token volume, spend, and in some cases images or videos generated.

That matters because model-routing arguments are rarely just about one dimension. Teams want to know what is winning, what is expensive, what is attracting production share, and whether image or video workloads behave differently from text workloads.

Once those dimensions can be exported, it becomes easier to ask questions like:

Those are much better questions than which screenshot looked impressive this morning?

The export endpoint is the operator hook that matters most

Vercel gives a concrete export path: https://vercel.com/api/ai/leaderboard-export.

It also says the current view can be downloaded as CSV and that the export data is cached for 24 hours. For models and labs, each row represents one entity's daily share of a single metric, with fields like requests, tokens, spend, imageCount, and videoCount.

That level of specificity matters because it makes the release testable. This is not a vague promise of transparency. It is a real data surface.

If you run a gateway, buy inference, or analyze model adoption, that is the difference between a marketing artifact and a dataset you can actually work with.

Open licensing changes who can reuse the rankings

Vercel also says the data is published under Creative Commons Attribution 4.0.

That is an underrated part of the announcement.

Open-ish access is not the same as openly reusable data. A clear license makes it easier for analysts, newsletters, vendors, and internal platform teams to cite, adapt, and remix the data without pretending the legal status is obvious.

The moment a leaderboard can travel as both an image and a dataset, it becomes easier for the surrounding ecosystem to build on it.

This is still platform evidence, not market evidence

It is worth keeping the claim narrow.

These rankings reflect AI Gateway usage, not the entire AI market. Apps are also described as opted-in. So nobody should confuse this with a neutral census of all production AI traffic everywhere.

But that does not make it unhelpful. It makes it scoped.

Scoped data is still useful when the scope is clear.

Teams should treat this as a benchmarking input, not a final answer

The right use for this release is comparison and context.

Teams can use it to benchmark:

That is enough to make the leaderboard meaningfully more operational.

Butler's take

I like this update because it pushes AI infrastructure discourse a little closer to evidence and a little further from vibes.

The charts themselves are fine. The real value is that Vercel is letting people take the data with them.

Once a leaderboard becomes exportable working data, it stops being only a dashboard feature and starts becoming a benchmarking loop.

Related coverage

AI Disclosure

This article was researched and drafted with AI assistance, then reviewed and edited for clarity, accuracy, and editorial quality.