Voice-agent developers
Replace the LLM step in an existing speech and telephony pipeline and benchmark a representative call set.
Independent tool overview
Mercury Voice: features, pricing and alternatives
Mercury Voice is an enterprise API language model tuned for the reasoning and tool-use step inside real-time voice agents; it does not itself provide the entire phone, speech and business workflow stack.
Visit the official Mercury Voice site ↗
- Product type
- Language model for voice-agent reasoning
- Access
- Enterprise Inception API; contact sales
- Context
- 128K tokens, vendor stated
- Pricing unit
- Input and output tokens
Overview
What Mercury Voice is
Inception describes Mercury Voice as a diffusion language model designed to respond quickly after a caller finishes speaking. Its September 2026 launch post says the model can reason, follow long instructions and call tools while reducing the pause before an agent starts talking. The published 320 ms median time to first answer token is a vendor-reported measurement on its production-style prompts, not a guarantee for a full call. Speech recognition, text-to-speech, telephony and tool services add their own latency.
The model is exposed through an OpenAI API-compatible endpoint for enterprise customers. This makes it a component a developer can put into an existing LiveKit, Pipecat, Vapi or Retell pipeline. Inception lists a 128K-token context window, up to 50K output tokens and low, medium and high reasoning settings. Teams should compare full turn latency, answer accuracy, tool reliability and cost on their own calls, especially long prompts and interruptions.
Use cases
Who Mercury Voice is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Call-center engineering teams
Test fast responses when agents must consult a knowledge base, update an order or transfer a call.
Platform builders
Offer a choice of reasoning effort and keep model selection separate from speech infrastructure.
Capabilities
Core Mercury Voice features
Low-latency output
Inception reports a 320 ms median first answer token on its test set; measure end-to-end call timing independently.
Tool calling
The model can invoke business tools for tasks such as lookups or booking inside a configured agent.
Reasoning control
Low, medium and high settings let builders trade additional thinking against response time.
Long instruction support
A 128K-token context accommodates prompts and retrieved conversation context, subject to cost and relevance.
Compatible API
The enterprise endpoint uses an OpenAI API-compatible interface for integration with existing frameworks.
Process
How the Mercury Voice workflow works
Step 1
Build a realistic call sample
Collect consented, anonymized examples including interruptions, vague requests, failures and escalation cases.
Step 2
Integrate the model
Request enterprise access and put the endpoint into the LLM slot of the existing voice stack.
Step 3
Bound actions
Give tools narrow permissions, confirmation rules and human handoff paths before live use.
Step 4
Measure the whole call
Track speech-to-text, model first-token, speech synthesis, correctness, transfers and actual token consumption.
Cost
Mercury Voice pricing and free plan
Inception publishes token rates for Mercury Voice and a launch discount. These are model costs; telephony, transcription, voice synthesis, orchestration and enterprise terms can add charges. The vendor’s per-minute illustration is an example profile, not a universal call price.
Standard model rate
$0.40/M input; $1.50/M output tokens
Published token rates for the Mercury Voice API.
- Enterprise access is through Inception sales
- External voice-stack costs are separate
Launch discount
$0.20/M input; $0.75/M output tokens
The announcement advertises a 50% launch discount.
- Discount duration was not specified on the launch page
- Confirm current rate in the customer agreement
Pricing checked . Check current pricing at the source ↗
Assessment
Mercury Voice strengths and limitations
Where it stands out
- Purpose-built for the latency-sensitive LLM stage of a live conversation.
- Reasoning levels allow tests across response quality and speed.
- OpenAI-compatible endpoint can simplify a trial in an existing agent framework.
What to consider
- Enterprise access requires contacting Inception; a self-serve Mercury Voice plan is not established.
- Vendor latency and benchmark numbers do not include all components of a deployed call.
- Teams must supply speech, telephony, permissions, monitoring and fallback behavior.
Compare
Mercury Voice alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Sales
Vapi
Compare a managed voice-agent platform if the team needs telephony and orchestration as well as a model.
Explore Vapi →Business Operations
Retell AI
Compare another phone-agent platform with call infrastructure and operational controls.
Explore Retell AI →Coding
Deepgram
Compare a specialist speech API for the transcription or synthesis stages rather than the language-model stage.
Explore Deepgram →Questions
Mercury Voice FAQs
Does Mercury Voice generate the spoken audio?
The launch describes an LLM for the reasoning step in a voice pipeline. It does not establish Mercury Voice as a standalone speech synthesizer or phone service.
How do I access it?
Inception says enterprise customers can use its API and should contact sales for access.
What does it cost?
The launch lists $0.40 per million input and $1.50 per million output tokens, with a 50% launch discount. Confirm the live rate and other voice-stack costs.
Is the reported latency guaranteed?
No. The 320 ms median is Inception’s test result for time to first answer token, not a service-level promise or whole-call measurement.
Bottom line
Our Mercury Voice verdict
Mercury Voice is worth a controlled trial for builders whose current voice agent pauses too long at the LLM step. Compare it against the current model using real call types, not only vendor charts, and include total stack cost and error handling in the decision.
Visit Mercury Voice website ↗Similar Miscellaneous Tools
View all Miscellaneous AI tools →
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.