Compute-Token

Pay for AI output, not idle GPU-hours

Compute-Token gives your teams one OpenAI-compatible endpoint to open-weight and frontier models, metered per million tokens. Tell us what you need to run and we will match you to the right capacity — shared or reserved, and in-region where it matters.

  • One OpenAI-compatible endpoint
  • Metered per million tokens
  • Shared or reserved capacity

Why buy compute by the token

A growing share of AI is bought as tokens through an API. Paying for output keeps your cost tied to the work actually done.

You pay for
GPU rental: Time on a GPU, busy or idle
Compute-Token: Tokens processed — input and output
When demand drops
GPU rental: The hourly rate keeps running
Compute-Token: Cost falls with usage
When demand grows
GPU rental: Provision and operate more GPUs
Compute-Token: Throughput scales behind the same endpoint
Capacity
GPU rental: One tenant per GPU
Compute-Token: Shared endpoints, or reserved capacity with guaranteed throughput
Pricing
GPU rental: One hourly rate per GPU type
Compute-Token: Tiered per-model pricing per million tokens

Model tiers

Every model in the catalogue sits in a tier, so you can match cost to the job — and switch models without changing your code.

Light

Fast, low-cost models for classification, extraction, routing and high-volume chat.

Typical models: Small open-weight models (about 1–8B parameters)

Standard

General-purpose assistants, retrieval over your documents, summarising and drafting.

Typical models: Mid-size open-weight models (about 14–70B parameters)

Premium

The strongest open and frontier models for complex writing, coding and analysis.

Typical models: Large open-weight and frontier models

Reasoning

Step-by-step reasoning for planning, mathematics, evaluation and agent workflows.

Typical models: Reasoning-tuned models

Each model is priced per million input and output tokens. Current prices are confirmed in your access offer.

From request to billing in four steps

  1. 1

    Choose your models

    Pick models by tier from the catalogue. New open-weight models are added as they are released.

  2. 2

    Get one endpoint

    Receive an OpenAI-compatible API key that works with the SDKs and tools you already use.

  3. 3

    Usage is metered

    Input and output tokens are counted per team, model and key, in real time.

  4. 4

    Budgets and billing

    Set token budgets per team and receive monthly statements with exportable usage data.

Every request routed by policy

Your rules decide where each request runs, so availability, cost and data location stay under control.

Burst when busy

When reserved capacity is full, overflow to approved external providers — only if your policy allows it.

Stay available

Requests fail over across data centres, so one outage does not stop your applications.

Budget guardrails

When a team reaches its token budget, fall back to an economy model instead of stopping work.

Data residency

Keep inference in-country and approve which models each team may use, with auditable records.

Built for how you use AI

Enterprises

One internal endpoint for every team, token budgets per department, and clear cost allocation.

Universities and training providers

Metered model access for courses, labs and research projects — without running GPU servers.

Builders and start-ups

Ship AI features on pay-as-you-go tokens, then move to reserved capacity as usage grows.

Frequently asked questions

What is a token?

A token is a small piece of text that a model reads or writes — about three-quarters of an English word on average. Usage is counted as input tokens (your prompt and context) and output tokens (the model's answer).

How is this different from renting GPUs?

With rental you pay for hours on a GPU whether it is busy or idle, and you run the serving stack yourself. With Compute-Token you pay for the tokens you use, and capacity, scaling and model updates are handled behind one endpoint.

Will it work with my existing tools?

Yes. The endpoint is OpenAI-compatible, so most SDKs, frameworks and tools work by changing the base URL and API key.

Where is my data processed?

Your access offer states where requests are processed. In-country deployment is available, and overflow to external providers happens only when your policy allows it.

What happens after I request access?

The AIKm team reviews your requirements and replies by email with the available models, capacity options and per-million-token prices.

Request access

Tell us what you plan to run. We reply by email with matching models, capacity options and prices.

Sign in to send an access request and follow its progress.

Sign in to request access