Hugging Face licensing covers two different things that share the word licence. The first is the commercial licence of the Hugging Face Hub, the hosting platform where users build, benchmark, share, version and deploy models, datasets and machine learning applications.[5] Hugging Face, Inc. sells that platform as subscriptions for personal accounts and organizations, and bills compute and extra storage on usage. The second is the licence a repository author attaches to a model, dataset or Space, which the Hub records in card metadata but does not grant on the author’s behalf.[6] A software asset manager normally has to track both: the seats and credits Hugging Face invoices, and the licence terms of the artifacts the organization downloads and deploys.
The commercial model has three layers. Subscriptions are flat fees: PRO is $9 per month for a personal account, Team is $20 per month per user and Enterprise is $50 per month per user on the pricing page.[1] Compute services (Spaces hardware, Inference Endpoints, Jobs, extra ZeroGPU time and Inference Providers) are billed on usage and are billed separately from the subscriptions.[3] Storage above the included amount is billed per terabyte.[4] Credits that come with a subscription are applied to compute before any pay-as-you-go charge.[3]
Editions
Hugging Face names its plans on three pages. The pricing page shows a PRO account at $9 per month, a Team plan at $20 per month per user and an Enterprise plan at $50 per month per user.[1] The plan comparison in the documentation adds a fourth organization tier, Enterprise Plus, priced “custom”, and lists a Free tier with no subscription fee.[2] Universities can buy Academia Hub, which starts at $10 per seat per month.[8]
| Plan | Account type | Price shown | Notable entitlements |
|---|---|---|---|
| Free | User or organization | None | Best-effort public storage, 100GB private storage[4] |
| PRO | Personal account | $9 per month[1] | Higher storage, $2 of monthly credits, higher ZeroGPU quota[1] |
| Team | Organization | $20 per user per month[1] | SSO, audit logs, resource groups, 12TB base public storage plus 1TB per seat[2] |
| Enterprise | Organization | From $50 per user per month[2] | SCIM invitation-based provisioning, service accounts, invoice billing, 200TB base public storage[2] |
| Enterprise Plus | Organization | Custom[2] | Managed SSO, network security, higher rate limits, customer-paper contracts[2] |
| Academia Hub | University | From $10 per seat per month[8] | 1 TB private and 1 TB public storage per seat[8] |
The Terms of Service also name other services that may be bought under order forms: Inference Providers, AutoTrain, Enterprise Support, the Infinity container, the Hardware Partner Program and Inference Endpoints.[5] The Terms define the Hugging Face Open-Source Libraries as the open-source projects published under the huggingface organization on GitHub, including Transformers, Datasets and Tokenizers, and say that open source licence terms are not replaced by the Terms.[5]
Metrics
Hugging Face does not publish a metric glossary. The catalog records the units that the plan, billing and quota pages use:
- Seat: a member of a Team or Enterprise organization. On renewal “the number of seats in your subscription will be updated to match the number of members of your organization”.[3]
- PRO account: a flat monthly fee for one personal account, payable only by credit card.[3]
- Private storage (TB) and Public storage add-on (TB): terabytes per month above the included storage.[4]
- Compute credits: dollar credits, $2 per PRO account and $2 per seat per month for Team and Enterprise.[10]
- Spaces hardware hour, Inference Endpoint instance hour and Jobs hardware hour: hours of hardware billed by the minute.
- ZeroGPU quota: minutes of daily GPU quota.
- Request quota per 5-minute window: requests per 5-minute window, applied per member.[9]
Counting / floors
Seats are not a counted licence position in the Oracle sense. They are billing quantities that follow membership at renewal, and the documentation says there is no need to update the subscribed seat count during the month or year because the subscription is a flat fee.[3] Subscriptions renew on the 1st of each month and the first month of a Team or PRO plan is prorated when the sign-up falls mid-month.[3]
Floors are few. Academia Hub requires at least 250 annual seats, and smaller groups are directed to Team.[8] Included storage scales with seats: Team and Enterprise organizations include 1TB of private storage per seat, so an organization with 40 members has 40TB of included private storage.[4] Request quotas are applied individually to each member, not shared across the organization.[9]
Licences on the Hub
Any repository can declare a licence in its card metadata, and the Hub lists the accepted identifiers, from apache-2.0 and mit through OpenRAIL, Llama and Gemma families to the Creative Commons set.[6] Authors can also gate a model or dataset, which makes users share their contact details and sometimes accept conditions before they can download the files.[7] The Terms of Service say that content carrying a customary licence notice stays under that licence when accessed, distributed or used, and that content downloaded from another user is used at the downloader’s own risk and subject to the terms accompanying it.[5] The article on model and dataset licences and gating treats these in detail.
Virtualization & partitioning
Not applicable. Hugging Face sells hosted services and does not publish processor, core or virtualization rules for the Hub. The Terms of Service state that the licence terms of open source items are not overridden by the Terms, so self-hosted use of the open-source libraries follows their own licences.[5]
Cloud / BYOL
Hosted compute runs on cloud providers. Inference Endpoints are priced per provider and instance type, with AWS, Azure and Google Cloud rows in the hourly tables.[1] Organizations can be billed through an AWS or Google Cloud marketplace subscription so that usage appears on the cloud bill, but only after Hugging Face approves the marketplace subscription.[3] For Inference Providers a customer can bring a provider key, in which case the provider bills the customer directly and the included monthly credits do not apply.[10] See Cloud marketplace billing and Custom Provider Key.
Programs
- Compute pay-as-you-go and credits: credits and usage billing for compute, invoiced at the start of each month.[3]
- Cloud marketplace billing: AWS and Google Cloud marketplace billing for organizations.[3]
- Community GPU grants: applications for help with GPU Space costs.[11]
- Storage grants for research teams and non-profits: case-by-case storage grants for high-impact open-source work.[4]
- Academia Hub program: the university subscription with a 250-seat minimum.[8]
- Enterprise Plus pilot: Enterprise Plus is the only plan for which the comparison table lists pilot availability.[2]
Contracts
The Terms of Service are a binding agreement between Hugging Face, Inc. and each user or customer. Supplemental terms, an Order Form, a Scope of Work or a Master Services Agreement can add to or override them, in that order of precedence.[5] The plan is billed in advance monthly, usage-based fees are billed as they go, and fees are non-refundable.[5] The Terms can change with 10 days of notice by posting.[5] See the Terms of Service and Content Policy article.
Out of scope
This article does not cover the licences of any specific model or dataset, the licences of the Hugging Face open-source libraries, AutoTrain, the Infinity container or the Hardware Partner Program, none of which publish price or metric pages that the catalog cites. It also does not cover HuggingChat or other consumer features, personal-data processing under the Privacy Policy, or export-control compliance for particular models. Practical steps such as inventorying organizations, reconciling seats at renewal and recording the licence of each downloaded artifact are practitioner commentary and are not Hugging Face requirements.