Membership¶
Every organization holds a membership level: Public, Pro, Max, Ultra or Steward, in that order. The level decides what the organization's membership includes: Marshall usage and workspace size per member, Base Load capacity, new managed clusters, the shared volume and CPU node allowances at the site, request budgets and support. Public is free and automatic. Pro, Max and Ultra are reached by paid spend in a month, a paid credit balance held, or a membership fee. Steward is by invitation only. An organization created before membership launched keeps the access it had then at every level: see grandfathered organizations.
The figures behind every level (what each path takes, how much each benefit includes) are your organization's own and live in the console, not on this page: where the figures are.
Levels¶
| Level | How an organization reaches it |
|---|---|
| Public | every organization, automatically and free |
| Pro | paid spend in a month, a paid balance held, or a membership fee, each at Pro's figure; or a grant |
| Max | the same paths at Max's figures, or a grant |
| Ultra | the same paths at Ultra's figures, or a grant |
| Steward | invitation only; no spend, balance or fee reaches it, and it cannot be bought |
An organization holds the highest level any path gives it. The paths do not add up: spend short of a level's figure and a balance short of its figure reach nothing together, however close each is.
A Public Research organization holds Pro (currently) for as long as it is one. National Compute can also grant a level, permanently or until a date; a granted level ends on that date and never soft-lands. When a Steward term ends, the organization holds Ultra for one more calendar month; a Steward level revoked for cause has no such month.
Paid spend¶
Spend in a calendar month (UTC) is the larger of the organization's paid deposits that month and its paid consumption that month. A level reached by spend applies the moment the month's spend reaches the level's figure, not at the month's end.
- Paid deposits are payments: card, wire, ACH and check, an invoice paid by wire included. A payment counts in full in the month it lands, also when part of it paid debt first.
- Paid consumption is usage paid from credits that came from a payment, locked credits included, plus usage drawn on an advance.
What counts as paid lists what does not.
Paid balance¶
The paid balance is the organization's Standard and Restricted credits that came from payments, locked credits included, less its debt. Each UTC day counts the lower of its opening and closing paid balance, never below zero. The balance path reaches a level in three ways:
- The month's average: the days' values summed over the month and divided by its days, at or above the level's figure.
- The running floor: the same sum so far, divided by all the days of the month. It only rises, so the level applies the day it crosses the figure.
- A streak: 7 consecutive completed days (currently) above the figure apply the level at once. A streak may span a month boundary.
Membership fee¶
A membership fee buys Pro, Max or Ultra directly, monthly or yearly. Any member of your organization buys, changes or cancels it on the console's Compare page: a card pays through a Stripe-hosted checkout, and Max and Ultra can be invoiced instead, due in 30 days. There is no fee verb on the token API or the MCP server; a person buys a level.
- The fee is a subscription, not credits: it adds no credits, is never paid from credits, writes no ledger row and never counts as spend (billing).
- The fee includes a set number of members. Each member beyond them adds a per-member fee, prorated as members join and leave, so every member gets the level's per-member allowances.
- An upgrade applies once its prorated charge is paid. A downgrade, or a switch between monthly and yearly, applies at the end of the paid period. One change at a time: an upgrade that also switches between monthly and yearly is refused.
- A cancelled fee keeps the level to the end of the paid period.
- A renewal not yet paid keeps the level for 7 days (currently) past the period's end, plus the 30 days to pay when the fee is invoiced. A cancelled fee gets no such days. A full refund or a lost dispute ends the paid period at once; a partial refund does not.
- A bought level ends with its paid period. It never soft-lands.
- An organization holds one fee at a time: a second purchase is refused while one is live, so change the level instead. An organization on a reserved arrangement arranges its level with its account team.
Hold and soft landing¶
A level earned by spend or balance holds through the end of the next calendar month, then steps down at most one level per month. Each month starts at the higher of the level the previous month's spend and balance reached, and one level below where the previous month ended. Qualifying again at any point in a month restores the level at once.
Ultra earned once, with no paid spend or balance after it:
| Month | Level |
|---|---|
| the month Ultra is earned | Ultra |
| the next month | Ultra, held |
| the month after | Max |
| the month after that | Pro |
| every month after | Public |
Only an earned level soft-lands. A bought level ends with its paid
period and a granted level on its end date; the organization then holds
whatever its other paths give it. The summary's
next_level and next_level_at name the
next scheduled change.
Notices¶
The Billing page shows a notice, and every member with an email address gets the same text by email:
| When | Notice | kind |
|---|---|---|
| an earned level steps down at the coming month boundary | from 7 days (currently) before the boundary, until it passes or the level is earned again | soft_landing |
| a granted level, Steward, or the Ultra month after Steward ends | from 30 days (currently) before it ends | grant_expiring |
| a grace period that lets an organization created after launch keep a capability below the level it needs ends; only once a requirement with grace periods is in force | from 30 days (currently) before it ends | exemption_ending |
| the level rises, except when a Public Research organization takes Pro | email only | none |
| a refund or reversal lowers a level already set | email only | none |
The summary's notices carries the same
notices for an agent.
What counts as paid¶
| Money | Counts as paid |
|---|---|
| A payment: card, wire, ACH or check | yes |
| Usage paid from credits that came from a payment | yes |
| Usage drawn on an advance | yes |
| Promo credits (sign-up credits, feedback rewards, data-sharing credits, promotions) and the usage they pay | no |
| A credit back and the usage it pays | no |
| Standard credits National Compute grants, and the usage they pay | no |
| Unpaid usage | no |
| The membership fee | no |
A deposit reversal or a Void of a payment counts against the month the payment landed in, whenever it is booked; a dispute that ends with the funds returned counts back in that month too. When that month already set the level, the level is recomputed from that month forward and can drop.
What each level includes¶
Some level requirements and benefits apply only once we turn them on: a new managed cluster needs Max, buying Base Load capacity needs Pro and a managed cluster to run on, and Max and Ultra workspaces are larger. The requirements bind organizations created after membership launched, and the table shows them that way: a grandfathered organization, one created before launch, has Base Load capacity and new managed clusters at every level. A requirement applies the moment it is turned on, or at the end of a grace period where it has one (notices), and a change to an allowance the moment it is made; advance notice of either is not guaranteed. Every cluster your organization already has keeps full access at any level.
| Public | Pro | Max | Ultra | Steward | |
|---|---|---|---|---|---|
| Burst capacity | yes | yes | yes | yes | yes |
| Base Load capacity, under its SLA | no | on your existing managed clusters | yes | yes | yes, with first access to new capacity |
| Managed compute | your existing clusters | your existing clusters | yes, new clusters too | yes | yes |
| Marshall usage per member, every week | included | included | more | unlimited (fair use) | unlimited (fair use) |
| Marshall workspace size | included | included | larger | larger | as Ultra |
| Resource monitoring and Grafana | yes | yes | yes | yes | yes |
| Shared volumes at the site | on request | on request | on request, an allowance included | on request, included, grows with your spend | as Ultra |
| CPU nodes at the site | on request, billed as used | on request, billed as used | on request, billed as used | on request, included, grows with your spend | as Ultra |
| Request budgets and the default max cluster size | default | default | higher | higher | higher |
| Support | priority email | enterprise, with a support SLA | custom, with solutions engineering |
- Managed compute is compute the platform operates for you: node health checks, automatic remediation, security patches, and the scheduling products on top. Unmanaged compute is GPU nodes handed to you whole; what runs on them is yours to operate. Every level includes unmanaged compute. A new managed cluster needs Max, once that requirement is on, unless your organization is grandfathered; a managed cluster your organization already has keeps everything it has at any level, including after a later change of level.
- Base Load capacity runs on a managed cluster your organization
has, and a new managed cluster needs Max, so an organization without
one gets Base Load with Max. On a managed cluster your organization
already has, Pro is enough. Every level of a grandfathered
organization includes it, and it carries the same
Base Load Capacity SLA wherever it is included. The
summary's
reserved_capacitysays whether your organization can buy it now, andreserved_requirednames the level it needs when it cannot. The Base Load book'scheckout_opensays whether self-serve buying is open where the cluster runs. - Marshall usage is model usage per member per week, the week starting Monday 00:00 UTC. Usage beyond the level's allowance is charged. Unlimited means up to a fair-use ceiling per member per week; usage beyond it is charged the same way.
- Marshall workspaces: one workspace per member at every level; the size grows at Max and Ultra, once that is on.
- Shared volumes and CPU nodes at the site are added on request at every level; the level sets the allowance that is not billed. Ultra's included amounts follow the previous month's paid spend.
- Request budgets: at Max and above, the token API and MCP request
budgets are twice the default (currently;
the summary's
rate_multiplier), and a Kubernetes cluster's max cluster size defaults larger; a size the platform set for the cluster stays as set. A cluster keeps the larger default size it had after a later change of level, for as long as it stays in your organization; a new cluster gets its level's default. - Support: the support SLA is separate from the Base Load Capacity SLA; its terms come from your account team.
A level never tears down what an organization already runs: clusters, volumes and nodes in place stay in place.
Grandfathered organizations¶
Organizations created before membership launched keep the access they had then, at every level, for good: their clusters (new managed clusters included), Base Load, shared volumes, CPU nodes and workspaces. Benefits that grow with a level still follow the level: the weekly Marshall allowance, the higher request budgets and default max cluster size, and the included shared volume and CPU node allowances. A change to a level's allowance, the Public Marshall allowance included, applies to them too. First access to new capacity stays Steward's.
An organization created after launch follows the level requirements.
National Compute can end one organization's grandfathering for abuse;
that organization then follows them too. A grandfathered
organization's Billing page says so, and the billing summary's
grandfathered reads true.
Resource monitoring and Grafana¶
Every level includes resource monitoring and the organization's own Grafana, readable by every member and embedded in each cluster's Monitoring tab in the console. Managed compute's dashboards read GPU and host telemetry from every node. Unmanaged compute has the Your nodes dashboard: each node's allocation state (active, provisioning or reclaiming), when it was granted and how long it has been held, and the cluster's declared and pending nodes. The platform sees nothing inside an unmanaged node, so telemetry from inside it is not on the board.
An agent reads the same series with
metrics_query and
lists the dashboards with dashboard_read.
Where the figures are¶
Each path's figure for each level and each benefit's amount are shown to your organization in the console and on the wire, never here:
- The console's Billing page shows the level, the path that gives it, the month's progress on each path, the date the level holds through, and any scheduled step down. The level also shows beside your organization's name.
- The Compare page (
/org/<organization>/billing/membership) lists all five levels with your organization's figures, the paths to each, and the membership fee's buy, change and cancel actions. - The billing summary
(
GET /api/billing/summary) and the MCP server'sbilling_readsummary view carry the same asmembership.