Compute
$0.04 / vCPU-hour
$0.02 / GB RAM-hour
Flexible resource allocation.
Usage-based billing.
Pay for the compute, memory, AI inference, and network resources your agents use.
$0.04 / vCPU-hour
$0.02 / GB RAM-hour
Flexible resource allocation.
Usage-based billing.
Your model.
Your token usage.
Model-specific token pricing. Input and output billed separately. Pay for the tokens consumed.
View model ratesFree ingress.
Measured egress.
Standard internet ingress is free in Mumbai. Outbound transfer follows published Mumbai tiers.
Explore transfer ratesStart with your allocation. Add tokens and transfer for the same scenario.
Allocated resources × runtime. A monthly estimate uses 730 hours.
Loading model prices…
Enter tokens per execution. Cached tokens are separate from uncached input. Executions multiply each category once.
Transfer is the total for your scenario. Private networking and separately configured networking services are not included.
Illustrative workloads to explore. These assumptions are examples, not fixed packages or usage guarantees.
Model rates per 1 million tokens. All prices use Mumbai. Inference mode and prompt size determine the applicable rate.
| Model | Category | Input / 1M | Output / 1M | Cached input / 1M | Calculator action |
|---|
Loading model pricing…
0 pricing options · Null or unavailable rates are never treated as free. Cache-write pricing, when supported, appears in the calculator. Admin billing uses these same published pricing options and saves the selected catalog version with its policy. Only configured options are listed; models without published prices and unsupported billing dimensions are omitted. Categories organize the catalog and are not performance guarantees.
Understand the assumptions behind your estimate.
Read the documentationThe estimate multiplies allocated vCPUs by $0.04 per hour and allocated memory in GB by $0.02 per hour, then adds the two. Fractional allocations are supported.
An idle workload can still hold allocated compute and memory. This calculator counts its entire runtime, including idle time. Closing or suspending a runtime and retaining its stored data are separate resource lifecycles.
The estimate uses the selected model, inference mode, and context tier in Mumbai. Enter token counts per execution; the number of executions is applied once to each supported token category. The admin billing settings select from this same versioned catalog. Your configured inference mode and context must match the Mumbai option; missing usage details hold invoice finalization.
Yes, where the selected model publishes different rates. Uncached input, output, cache reads, and cache writes are separate categories. Do not include a cached token again in the uncached input field. An unavailable rate does not mean free.
Standard internet ingress is $0 per GB in Mumbai. Other transfer paths, private connectivity, and separately configured networking services are outside this calculator.
Outbound traffic uses Mumbai’s published tiers. No free allowance is assumed. Apply an allowance only after confirming that your account qualifies and that it has not been consumed elsewhere.
Yes. Select the 730-hour preset for monthly compute. Tokens and transfer are totals for your scenario and are not automatically multiplied by runtime. The Always-On example explicitly assumes one execution for each of 730 hours.
Prices are maintained in the Deploymate admin dashboard and loaded when this page opens. Admin changes appear on your next refresh. The catalog version and date identify the displayed prices. If pricing is unavailable, we show that instead of using old model rates.
Start with an estimate. Build on your terms.