Costs

Every call and chat carries two costs. Provider cost is what your LLM, STT, TTS, realtime and avatar usage cost at the providers’ rates; your providers bill it to your own keys and it never passes through Talqing. Platform fee is Talqing’s per-minute (calls) or per-answered-message (chats) charge, and is the only thing debited from your credit balance. total_charge is the sum, for seeing a call’s full economics in one figure. Rates, credits, the ledger and what happens at zero balance are on pricing and credits.

How the platform fee is computed

A call’s fee is its channel’s per-minute rate times its duration, prorated per second. Avatar time on a video call is provider cost, not part of the fee.
A chat’s fee is per answered message: a message whose turn finished done. Superseded (canceled) and failed (error) turns are free.
The fee is waived on a failed call; provider cost still applies. Which close reasons count as failed is listed on close reasons. transferred is not a failure and is charged.

Voicemail is an answered call. An outbound call that reaches voicemail connects and is billed. With voicemail detection the agent hangs up after its message (voicemail); without it the agent keeps talking to the machine.

The five metered kinds

Each provider charge is priced from the catalog entry of the exact model that ran (GET /v1/catalog).

Kind	Priced on
`llm`	Input, cached input and output tokens per 1M. Cached input uses the model’s cached rate, or the full input rate if it has none.
`stt`	Audio seconds and/or tokens, as the provider sells it.
`tts`	Characters, audio seconds and/or tokens, as the provider sells it.
`realtime`	Text and audio tokens in (cached and uncached) and out, plus a per-session-minute rate for providers that sell by the minute.
`avatar`	Wall-clock time the avatar is in the call, idle included, per second.

A call has either llm + stt + tts or realtime, never both; avatar sits beside either. LLM lines carry purpose (conversation, or analysis for post-call analysis) and priority (the line ran in the priority lane at its higher rate). Noise cancellation is not metered.

One call’s cost

GET /v1/calls/{session_id} returns cost and session.billing_status.

curl https://api.talqing.com/v1/calls/9c1f6ad2-3b47-4e58-9a2c-7f0d51e8c4b1 \
  -H "Authorization: Bearer $TALQING_API_KEY"

const call = await talqing.calls.get({ session_id: "9c1f6ad2-3b47-4e58-9a2c-7f0d51e8c4b1" });
if (call.cost) {
  console.log(call.cost.provider_cost, call.cost.platform_fee, call.cost.total_charge);
  for (const line of call.cost.pricing_snapshot?.llm ?? []) {
    console.log(line.provider, line.model, line.purpose, line.priority, line.cost);
  }
}

call = talqing.calls.get("9c1f6ad2-3b47-4e58-9a2c-7f0d51e8c4b1")
if call["cost"]:
    cost = call["cost"]
    print(cost["provider_cost"], cost["platform_fee"], cost["total_charge"])
    for line in cost["pricing_snapshot"]["llm"]:
        print(line["provider"], line["model"], line["purpose"], line["cost"])

Field	Meaning
`provider_cost`	Sum of every priced line.
`platform_fee`	Talqing’s fee; `0` when waived.
`total_charge`	The two added.
`duration_s`	Duration the fee was computed over.
`pricing_snapshot`	The breakdown frozen at settlement: `llm`, `stt`, `tts`, `realtime`, `avatar` line arrays (each line with quantities, `cost` and the `rates` applied), plus `platform_fee_per_minute` on calls or `platform_fee_per_message` and `answered_messages` on chats. Never re-priced from today’s catalog.

`billing_status`	Meaning
`pending`	Not priced yet; `cost` is `null`. Normal while a call runs. On a call that ended long ago, the process died before pricing and nothing will price it later.
`computed`	Priced; `cost` is set.
`unpriceable`	The call used something the catalog cannot price. It is not charged; `cost` is `null` and the reason is in observability `cost.unpriceable_reasons`. Contact support with the call id.

A chat is priced as it goes: GET /v1/chats/{chat_id} returns cost so far while it is open, final once it ends.

Workspace cost

GET /v1/observability (up to seven days; see observability) returns a cost section:

Field	Meaning
`priced_sessions`	Sessions these figures cover. Every figure is over this subset.
`provider_cost`, `platform_fee`, `total_cost`	Range totals.
`llm_cost`, `stt_cost`, `tts_cost`, `realtime_cost`, `avatar_cost`	Provider cost by kind.
`analysis_cost`	The part of `llm_cost` spent on post-call analysis.
`priority_cost`	The part of `llm_cost` run in the priority lane.
`llm_input_tokens`, `llm_cached_input_tokens`, `cache_hit_ratio`	How much LLM input hit the provider’s prompt cache.
`lines`	One row per kind, provider and model, with `cost` and `share`.
`by_provider`	Spend per provider as a per-bucket series.
`unpriceable_reasons`	`{message, sessions}` per distinct reason.

GET /v1/calls/stats gives total_spend and avg_cost_per_call over priced calls for a 30-day default window. Agent task runs are not sessions and are in none of these totals; their cost is on each run (running tasks) and, for email drafting, on the batch.

In the dashboard

The call page’s Cost section shows where provider cost went, one row per priced line with quantities and rates (priority and analysis tagged), then provider cost, the platform fee with its formula or waived — the call failed, and the total. Amounts show six decimal places.

Pricing and credits

Rates, credits, the ledger, zero balance.

Provider keys

Which providers bill your keys.

Get started

Build

Guides

How the platform fee is computed

The five metered kinds

One call’s cost

Workspace cost

In the dashboard

Pricing and credits

Provider keys

​How the platform fee is computed

​The five metered kinds

​One call’s cost

​Workspace cost

​In the dashboard

Pricing and credits

Provider keys

How the platform fee is computed

The five metered kinds

One call’s cost

Workspace cost

In the dashboard