Pricing
Two service lines, transparent billing models — per GPU-second, per token, and per node-hour — each with enterprise agreements for regulated workloads. Final terms are agreed in your service agreement.
GPU plans
Dedicated, isolated GPU infrastructure in Japan for training and inference. Billing depends on the plan you choose — final configuration and terms depend on your workload.
For research labs and ML teams getting started on dedicated capacity.
GPU usage metered per second — environment included
For production training and inference workloads that need isolation and support.
Reserved capacity billed per node-hour
For banks and regulated enterprises with bespoke compliance and architecture requirements.
Custom capacity and terms, agreed individually
Billing details are confirmed in your service agreement.
LLM services
Two ways to run models — dedicated deployments billed per GPU-second, or shared endpoints billed per token — all through one OpenAI-compatible API.
Open-weight models
Open-weight models served from Tara Cloud's own GPU infrastructure in Japan — every prompt and completion stays in-country.
Per token on shared endpoints — or per GPU-second on dedicated deployments
Frontier models
The official APIs, on the same terms — with unified JPY billing through Tara Cloud. OpenAI and Anthropic's official models are served through each vendor's official API, so their inference may run outside Japan.
Per token, with one monthly JPY invoice
Tell us about your workloads and we will prepare a tailored proposal — usually within two business days.