537 models are available to the enterprise. Twenty-three are carrying work at full scale.
Everything underneath the deployments, in one table. Which model, from which provider, at what context and price, and how many confirmed enterprise deployments in the register are actually running on it. Sort it, filter it, quote it.
Three providers carry half the confirmed deployments in the register
Model choice is the most discussed decision in enterprise AI and the most concentrated one. The long tail is long and almost entirely unconfirmed.
Register edition 2026.4 · compiled 19 Oct 2026
A catalogue is not a market
Every provider publishes a model list. None of them publishes what is running on it inside an enterprise. That second column is the only one that changes a decision, so it is the one this page is built around.
Availability is not adoption
Five hundred and thirty-seven models can be called from an enterprise account today. Three hundred and sixty-three appear in at least one confirmed deployment in the register. Twenty-three appear in a deployment an organization has confirmed as running at full scale. Each of those is a different number and they get quoted as if they were one.
The concentration is sharper than the discourse
The top three providers account for roughly half of every confirmed deployment the register holds. Benchmark leadership moves monthly. What is actually in production moves at the speed of a procurement cycle, which is why the table below looks less dynamic than the leaderboards do.
Price is the least decisive column
Token price is the easiest thing to compare and the smallest line in a deployment budget. Across the enquiries behind these records, model cost is rarely the reason a program stopped. Integration, evaluation and the work of changing a process are.
The free tier is a trial, not a deployment
Free variants carry a large share of traffic and almost none of the confirmed deployments. Where a free variant does appear in a confirmed record, it is nearly always alongside a paid model doing the work that matters.
Every model we index, and what is running on it
Sort any column. Filter by modality, provider or the deepest stage the register can confirm. Context and price are the provider's published figures at the compilation date. Deployments are counts from the register, not traffic, downloads or benchmark scores.
| Model↓ | Provider↓ | Released↓ | Context↓ | Input $/M↓ | Output $/M↓ | Modality↓ | Deployments↓ | Deepest stage↓ |
|---|
The table opens newest first, and the deployments column at the top of it is almost entirely empty. That is not a gap in the data. A model released this month cannot yet appear in a confirmed enterprise deployment, because confirmation takes a procurement cycle and then an answer from the organization running it. Sort by deployments to see what is actually carrying work, and read the two orderings against each other. The distance between them is the single most useful thing on this page.
Context lengths and prices are the provider's own published figures as at 19 October 2026 and move without notice, so treat them as indicative and check before you contract. Deployment counts are confirmed records in the register for that compilation date. A model with zero is not a judgement on the model. It means no claim naming it has yet been confirmed by the organization named in it.
Where the confirmed deployments actually sit
Confirmed deployments in the register, by the provider of the model carrying the work. Ten providers out of sixty-one.
The tail is not the story, but it is not nothing either. Fifty-one providers sit outside this chart and hold a real minority of confirmed deployments between them, heavily weighted to embedding, transcription and image work rather than to the general models that get argued about. Specialist models are being deployed quietly while the general ones are being debated loudly.
What we record about a model, and what we refuse to
This page is a record, not a league table. The Index gives no score and no grade, and that rule applies to models exactly as it applies to vendors.
We record the model named in a confirmed claim
When an organization confirms a deployment, we ask which model carries it. Where they answer, the deployment is attributed here. Where they decline, the deployment still counts in the register and does not appear in this table.
We do not benchmark and we do not rank
No score, no grade, no ordering by quality. There are good public benchmarks and we are not one of them. Sorting this table by deployments tells you what is being used, which is a different question from what is best.
Specifications are the provider's, attributed and dated
Context length and price are taken from the provider's published material on the compilation date and carried unchanged. We do not adjust, normalise or estimate them, and we do not negotiate on your behalf.
Records lapse rather than persist
A confirmed deployment lapses to unconfirmed twelve months after it was answered. Model attributions lapse with them, which is why counts here fall as well as rise between editions.
Two ways to engage us, and two things included with either
Everything rests on the same register. We do not resell inference, take referral fees, or accept sponsorship from anyone who trains or serves a model.
Expert guidance
A standing line to the people who maintain the register. Bring a model selection, a multi-model architecture, a renewal, or a program that has stopped moving.
- Inquiry calls and written responses inside 48 hours
- Shortlist reviews and proposal reads before you sign
- Bought as sessions, not seats, with register access included
Executive workshops
Your own decision, worked through by the people who have to own it, against a picture of what is actually deployed rather than what is currently topping a leaderboard.
- A function, the leadership team, or the management layer
- Our maturity assessment, frameworks and canvas, brought to the room
- Moves chosen, owners named and a written record in five working days
Model review
Commissioned. Every confirmed deployment in the register running on the models you are choosing between, and what the organizations said about them.
Register access
Included with guidance. The full table with every field, filtered any way you need, exportable, with alerts when an attribution lands or lapses.
Bring us the model decision in front of you
A selection between two or three, an architecture you are about to commit to, or a model you are being told to standardise on. We will tell you what the register can answer, what it cannot, and which of our services fits. If none of them does, we will say that.
What people ask about this table
No. The Index gives no score and no grade to anything, models included. The table defaults to sorting by confirmed deployments because that is the column most buyers come here for, but a high count means widely used, not best. Those are different claims and we only make the first one.
From the register. When an organization confirms a vendor's claim, we ask which model carries the deployment. Where they answer, it is attributed here. These are not API traffic figures, token counts or download numbers, and they are not comparable to the usage charts the inference marketplaces publish.
They are the provider's published figures as at the compilation date printed on the page. Model pricing changes faster than any other field here, often without announcement, so treat these as indicative for comparison and confirm the number with the provider before it goes into a contract or a business case.
Because no claim naming it has yet been confirmed by the organization named in that claim. A recent release will almost always sit at zero for a quarter or two. So will a model used heavily for internal work that no vendor has published a customer story about. Zero is a statement about the evidence, not about the model.
Yes, on the same terms, with one honest limitation. An organization running an open-weight model on its own infrastructure often has no vendor attached to publish a claim in the first place, so open-weight deployment is systematically under-represented here. We say so rather than letting the gap read as absence.
Yes. Register access gives you the whole table with every field, filtered to the providers, modalities and industries you care about, exportable, with alerts when an attribution lands or lapses. A commissioned model review goes further and takes the specific models in front of you through every confirmed record that names them.
Yes, and it is free. Request it and we will open a record, capture the published specification and attribute any confirmed deployments that name it. That is usually complete inside ten working days.
The catalogue is refreshed continuously as providers publish. The deployment counts are rebuilt each quarter with the register edition, which is why this page carries an edition number. Quoting it without that edition and date attached will date the number badly inside a month.
Before you standardise on a model
Check what the register holds on the models in front of you, commission a model review, or bring the decision into a room with your own leadership team.