Last verified
ENTERPRISE ONLY50% OFF AT LAUNCHVOICE AGENTS128K CONTEXTTEXT IN, TEXT OUT

Mercury Voice API Pricing

Mercury Voice is Inception's diffusion LLM tuned for voice agents: the language model in a speech pipeline, not a speech model, so it takes text and returns text. It has been generally available to enterprise customers since September 29, 2026, enabled per organisation through Inception's sales team. The list price is $0.40/M input, $0.04/M cached input and $1.50/M output; at launch it is 50% off, $0.20 / $0.02 / $0.75, with no end date published.

Input - per 1M tokens
$0.40/M
Charged now $0.20/M - 50% off list
Output - per 1M tokens
$1.50/M
Charged now $0.75/M - 50% off list
Cached input - per 1M tokens
$0.04/M
Charged now $0.02/M - 50% off -90%
Effective - agentic blend
$0.216/M
92/8 split - 82% cache
§ 01 / TERMINAL

Run the numbers.

Calculator pre-loaded with Mercury Voice's LIST rates. While the 50% launch discount runs, halve every result for the bill an enterprise account gets today. A voice turn re-sends the system prompt and the conversation so far, so most of each turn is input, and much of it can be cached.

$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
Open full calculator (all models · share URL · CSV) →
§ 02 / SCENARIOS

Real-world presets.

VOICE AGENT

Live call turn

$0.00108/turn
6,000 in - 300 out~92,695 units/$100
DRIVE-THRU

Order with modifications

$0.00644/order
40,000 in - 1,500 out~15,523 units/$100
COLLECTIONS

Payment-plan call

$0.019/call
120,000 in - 4,000 out~5,383 units/$100
SUPPORT

Phone ticket with tools

$0.010/call
60,000 in - 2,500 out~9,962 units/$100
§ 03 / TOKENIZER

Paste text. See tokens. See cost.

What's a token →
Estimate · inception-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type

This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.

Characters 436
Words 70
Tokens (estimated) 113 tokens
Cost as input · uncached $0.00005 USD
Cost as output · uncached $0.00017 USD
Cost as cached input $0.000005 USD
§ 04 / SHELF

Up against the shelf.

All models →
Model Input /M Output /M Effective blended Context Best for
Mercury Voice Current $0.40 cache $0.04 $1.50 $0.216 agentic 92/8 128K Low-latency voice agents on an enterprise account
Mercury 3 $0.20 cache $0.02 $0.70 $0.104 −2.1× · promo 1M Self-serve voice and agents, 1M window, half the list price
Mercury 2.5 $0.20 cache $0.02 $0.75 $0.108 deprecated · −2.0× 260K The previous generation, closed to new users
GPT-6 Luna $0.10 cache $0.01 $0.50 $0.0682 −3.2× 1.05M OpenAI's fast tier, the one Inception benchmarks against
Gemini 3.5 Flash-Lite $0.30 cache $0.03 $2.50 $0.272 +26% 1M Google's cheapest multimodal tier
GLM-5.3-Flash $0.15 cache $0.03 $0.50 $0.0875 −2.5× 1M Cheap fast tier from Z.ai
Claude Haiku 5.5 $0.10 cache $0.01 $0.50 $0.0682 −3.2× 1M Anthropic's fast tier with vision

Frequently asked.

What Mercury Voice costs per token and per minute, how to get it, and when Mercury 3 is the better buy for a voice agent.

Q · 01 How much does Mercury Voice cost? +
The list price is $0.40/M input, $0.04/M cached input and $1.50/M output, $0.22/M on our 92/8 blend with 82% cache hits. At launch Inception takes 50% off: $0.20 / $0.02 / $0.75. The cached rate is printed on the developer table; the launch post gives only input and output. No end date for the discount is published.
Q · 02 Is it a speech model? +
No. It is a text model built for the language step of a voice agent: speech-to-text produces the caller's words, Mercury Voice reasons, calls tools and writes the reply, and text-to-speech speaks it. Both the speech steps are billed separately by whichever vendor runs them. Inception says it drops into LiveKit, Pipecat, Vapi, Retell or a stack of your own through its OpenAI-compatible API.
Q · 03 How do I get access? +
Through Inception's sales team. The documentation says Mercury Voice is available to enterprise customers and that you contact [email protected] to enable it for your organisation; no public model id is published, and it is not on OpenRouter. That is why it is kept off this site's public leaderboards even though the rate is public.
Q · 04 What does a minute of conversation cost? +
Inception's own figure is about $0.009 per minute at the launch price for what it calls a typical voice-agent profile, and it compares that with $0.045 for GPT-4.1. It does not publish the token profile behind the number, so treat it as the vendor's estimate. The calculator on this page lets you set your own turn size; at list, double whatever the launch price gives.
Q · 05 Mercury Voice or Mercury 3? +
Mercury 3 is the self-serve choice, and it is what Inception's own voice quickstart and cookbooks use. It lists at $0.20 input and $0.70 output, below this model's card on both, with a 1M window against 128K. Mercury Voice is tuned for time to first answer token rather than end-to-end latency; Inception reports a 320 ms median and 750 ms at p95 on production voice prompts. Pay the premium when that first-word latency is the binding constraint and you have an enterprise contract.
Q · 06 How fast is it? +
Inception reports a median time to first answer token of 320 ms (p95 750 ms) on real customer-service prompts, 5.9 times faster than GPT-6 Luna without reasoning, and says it beats Gemma 4 31B, GPT-6 Luna, GLM-5.3-Flash and Qwen3.5-397B on a composite of agentic and conversational benchmarks. Its models page cites under 170 ms from one customer's production workload. All of these are the vendor's measurements, not ours.
Y
Reviewed by Yaroslav Vikhariev Founder · AI//COST · Prices checked by hand against docs.inceptionlabs.ai · Last verified Oct 21, 2026
Methodology Report a correction More by Y.V.