infinitetokens · PaaS

Infinite tokens for a finite time.

An AI platform for scoped model engineering and serving. Bring a scope of work, training or fine-tuning a model, RLHF and evaluation campaigns, synthetic data, a product launch or a research programme, and we reserve training clusters and inference for it on your own endpoint for the length of the work. You are billed for the platform and the term, never per token.

Talk to us Need seats instead? See Teams

01 — The shape of an agreement

From scope of work to a live endpoint.

  1. Scope of workIt starts with a bounded scope of work.
  2. The shared fleetOur platform: open-weight models on GPUs we operate.
  3. Reserved for youWe reserve a slice of it for your work alone.
  4. Your endpointIt is served on your own endpoint, under your company's name.
  5. Infinite tokensInside the scope, tokens are not counted.
  6. A finite timeThe term is fixed: 3, 6, 9 or 12 months.
  7. Review and extendAt the end, extend the term or hand the capacity back.
  8. Bounded work, unbounded tokensIf the work is bounded and the tokens are not, it fits.

02 — What you get

Capacity that is yours for the term.

  • Your own address

    <company>.infinitetokens.ai, an endpoint named for your organisation.

  • Reserved platform capacity

    Capacity held for you, not shared out when demand rises elsewhere.

  • Your choice of open models

    The open-weight models your work needs, chosen with you.

  • Isolation

    Your workload is separated from other customers.

03 — How it works

From requirements to a live endpoint.

  1. Size the workThe scope, the models, the concurrency and how the work will run through the day.
  2. Reserve the platformWe set inference capacity aside for the agreed term.
  3. Go liveYour endpoint is created on your name and connected to your pipelines.
  4. Run, then renew or changeAdjust at the end of the term as the work changes.

04 — Scopes we take on

Bounded work, unbounded tokens.

Any organisation can use it. If you know what the work is and how long it runs, the tokens should not be the thing you ration. These are examples, not a definition.

  • For example, improving a model: RLHF, fine-tuning and distillation data.
  • For example, evaluation and red-teaming campaigns.
  • For example, agent and product launches that need guaranteed headroom.
  • For example, bulk processing: document backlogs, code migrations and data labelling.
  • For example, research programmes that run many experiments in parallel.

If the work is bounded and the tokens are not, it fits.

05 — Commercials

Agreed per contract.

  • 3, 6, 9 or 12 monthsThe term of the reservation.
  • Set per agreementScope, capacity, models, term and endpoint are agreed with you.
  • No token meteringBilled for the platform and the term, not by what you send.

Talk to us

Tell us what you need to run.

Every agreement is shaped to the workload: seats, sessions, capacity, term and location.

Talk to us hello@mirastacklabs.ai