The credit-token model
Managed usage is measured in credit tokens, where 1 token = 1 US cent of managed cost. For BRAHMA, that managed cost is the amortized cost of serving the completion on your self-hosted GPU — compute time, memory, and the surrounding platform, spread across the tokens generated. As a reference point, 100 credit tokens ≈ $1.00.Rates
BRAHMA is priced with separate input and output rates, following the same convention as other model providers’ pricing pages. Prompt (input) and completion (output) tokens are metered separately — matching theprompt_tokens / completion_tokens split in the usage object — and each dimension is billed at its own rate:
Why output costs more. Generating a token is decode-bound: the GPU produces output tokens one at a time, so completion tokens dominate the serving cost of self-hosted H100 infrastructure. Reading your prompt is prefill — processed in large parallel batches and vastly faster on-GPU — so input tokens cost a small fraction of output tokens. The $2.00 input rate is a floored price that covers proxy and metering overhead; the derived infrastructure cost of input alone is lower. Both rates reflect the cost basis of serving BRAHMA on self-hosted GPU infrastructure and are reviewed periodically.
As a rule of thumb, a typical voice-call turn consumes far more prompt tokens (conversation history and instructions) than completion tokens, so the blended per-call cost is substantially lower than under the previous flat rate. A call with roughly 2,000–4,000 total tokens typically works out to a few cents.
Every BRAHMA chat-completion — streaming or not — records the prompt and completion tokens it consumed against the organization that owns the API key, and converts them to managed credit-token cost at the per-dimension rates above. New organizations start with a free credit-token allowance; consumption draws it down.
Credit-token metering is a managed-deployment concept. On OSS / on-premises self-host deployments where billing is off, BRAHMA usage is unlimited, the billing routes return
404, and token counts are recorded for visibility only.Reading your usage
Retrieve your organization’s current BRAHMA token consumption and derived cost from the Token Usage endpoint — it returnsprompt_tokens, completion_tokens, and total_tokens for the organization scoped to your API key.
For the full picture of credit balances, free-tier allowances, and how the credit-token model applies across managed usage, see the Billing & Credits reference.