Theme

Pricing overview ​

API requests use credits. Check the app's Billing panel for speech and audio rates, and GET /v1/models for LLM rates.

What is metered ​

ServiceTypical metricNotes
LLMInput and output tokensModel-specific rates from GET /v1/models; fractional credits accumulate per account
TTSCharactersGenerated text length; realtime checks balance when the session starts
STTAudio durationAll uploads use /v1/stt and return asynchronous jobs
Voice IsolatorAudio durationCredits rounded up per file; optional Sidon restoration has no separate surcharge
Meeting AIRecording durationCredits rounded up per meeting; notes regeneration is charged again

LLM, Voice Isolator, and Meeting AI return 402 insufficient_credits for insufficient account balance and 402 api_key_credit_limit when a key limit is exceeded.

LLM billing ​

Read current prices and the credits-per-USD conversion from GET /v1/models. Prices distinguish input and output and are expressed in USD cents per million tokens. Each admitted request keeps its price version even if current prices change.

Before generation, VoiceLab reserves a conservative input budget plus the maximum output budget. The available balance and API-key credit limit must cover this reservation, which can exceed the eventual charge. Reported usage settles the charge and releases the unused reservation.

Input includes conversation history, function definitions, and tool results. Output includes function arguments and reported reasoning tokens, counted once within completion tokens. Cached input uses the normal input rate. Fractions accumulate per account at 1/100,000,000 of a credit; a small request may debit zero whole credits while still accruing usage.

Model discovery and request-status lookups are free. Completions use developer credits even on an unlimited platform plan. Ask's JWT routes remain uncharged.

Read GET /v1/llm/requests/{id} for the exact credit units, whole credits debited, token counts, and price version. A failed request can still be charged when usage is known. Missing final usage can block further LLM generations pending support reconciliation. Do not recover by submitting the same generation under a new idempotency key.

Voice Isolator billing ​

Voice Isolator charges by duration at the configured per-minute rate, rounded up per file.

Credits are held when a job is accepted and settled once model processing succeeds. Terminal failure releases the hold. A batch reserves credits atomically; an invalid or unaffordable batch admits no jobs. Retrying the same upload with the same idempotency key and content does not create another hold.

Export conversions do not rerun the models or add credits. Hiding a job does not cancel accepted work or refund it.

Voice Isolator checks a key's credit limit against recorded billable API usage plus that key's pending Voice Isolator reservations. 402 insufficient_credits means the account balance is too low; 402 api_key_credit_limit means the key limit would be exceeded.

Successful developer jobs appear under voice_enhancer in API usage analytics.

Meeting AI billing ​

Meeting AI charges by recording duration at the configured per-minute rate, rounded up per meeting. Credits are held when an upload is accepted and settled once processing succeeds. Processing failure or deletion before completion releases the hold. Retrying the same upload with the same idempotency key and content does not create another hold.

Regenerating notes after a notes failure reserves and charges the meeting's full duration again. Polling, listing, renaming and deleting are free.

A key's credit limit counts recorded billable API usage plus that key's pending Meeting AI, Voice Isolator, and LLM reservations. Successful developer meetings appear under meeting_ai in API usage analytics.

Where to monitor usage ​

Enterprise needs ​

For volume pricing or an SLA, contact VoiceLab.

Next ​