◆
Tokens priced against the leading cloud
Input and output tokens on the general model cost 30% under the leading cloud’s model service for the same model. The comparison is published, so nobody has to take it on trust.
An agent is a loop: it calls a model, calls a tool, reads the result and goes round again. Every turn is a charge somewhere. On Compute the model, the machines and the state an agent keeps sit on one bill, priced below what the leading cloud charges for the same thing, and you pay for the turns that happened rather than the capacity you guessed at.
Used by startups shipping their first agent and by enterprises running fleets of them.
What this looks like in practice
Input and output tokens on the general model cost 30% under the leading cloud’s model service for the same model. The comparison is published, so nobody has to take it on trust.
Tool servers and GPU jobs are billed by the second. An agent that is idle overnight costs nothing overnight.
Memory and history live in CosmoBase, artefacts in Cloud Services, code in CodeGraph. One account, one set of permissions, one invoice.
No reserved throughput and no minimum term. Agent workloads change shape every month, and the bill is allowed to change with them.
Most agent projects stall at the same point: the demo works, and then someone multiplies the token bill by the number of customers. The answer is rarely a cleverer prompt. It is paying less for the same model and only for what actually ran.
Whether an agent becomes a product depends on one figure: what a completed task costs. On Compute that figure is lower because the tokens cost less than on the leading cloud and the machines only bill while they run. It is also easier to know, because every part of the task lands on the same bill with a published price beside the leading cloud’s.
Function endpoints and tool servers on Compute Instances, sized for the work and billed by the second.
Conversation history and agent state in CosmoBase, with two databases included before you pay anything.
Coding agents open changes in CodeGraph, run CI and leave a reviewable record of what they did.
Replay a large set of conversations on GPUs for an hour, then stop paying for them.
Compared with the leading cloud's own published on-demand prices, US East. Every comparison.
Open-weight models through one API, billed per input and output token. The current list and every price are on the pricing page, with the leading cloud’s price for the same model beside it where one exists.
No. The price on the page is the price at any volume, with no reservation, no minimum and no term. A startup running its first agent and an enterprise running hundreds pay the same rate.
Usually the change is an endpoint and a key. Move the model calls first, compare a week of cost and behaviour against what you had, and move the tools and state when you are satisfied.
Where data lives and who may reach it are configured properties of your account. If your sector requires a specific attestation, ask us directly and we will tell you plainly what we hold today.
Host the landing pages, attribution pipelines and onboarding flows that bring customers in, and spend less on the infrastructure than you would on the leading cloud.
Run campaign sites, personalisation, content pipelines and marketing data on one platform that costs less than the leading cloud for the same machines and models.
Run the forms, enrichment jobs, scoring models and routing that turn interest into leads, for less than the same infrastructure costs on the leading cloud.
Run CRMs, sales data, call analysis and AI assistants on one platform, priced under the leading cloud for the same machines, databases and models.
Host creator platforms, media processing, storefronts and AI tools on infrastructure that bills by the second and costs less than the leading cloud for the same machines.
Run your product, its databases, CI and AI features on one platform that costs less than the leading cloud for the same machines, databases and models.
Begin with one workload. Keep what you have running. Decide whether to continue once you have seen it carry something real.
No migration required to begin. Nothing to tear down if you stop.