Mercury Voice
Inception's diffusion language model tuned for voice agents, returning a first answer token in under 320ms median while reasoning and calling tools.
Overview
Inception launched Mercury Voice, a diffusion LLM (dLLM) tuned to power voice agents that reason, call tools, and follow long system prompts while keeping latency low enough for natural conversation. On a test set of real customer-service prompts it returns its first answer token in under 320 milliseconds median, with reported time-to-first-token as low as 170ms, and beats models including GPT-6 Luna and Gemma 4 31B on a suite of agentic and voice benchmarks. Unlike autoregressive LLMs that generate text one token at a time, Mercury's diffusion architecture generates tokens in parallel for 5-7x higher throughput and up to 70% lower cost per task. Mercury Voice is available to enterprise customers through the OpenAI-compatible Inception API and integrates with LiveKit, Pipecat, Vapi, and Retell, or a custom voice stack, priced at $0.40 per million input tokens and $1.50 per million output tokens, 50% off at launch.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Sub-320ms median time-to-first-answer-token on real customer-service prompts
- Diffusion-based parallel token generation for 5-7x higher throughput than autoregressive LLMs
- Reasons, calls tools, and follows long system prompts within a voice-latency budget
- Beats GPT-6 Luna and Gemma 4 31B on agentic and voice benchmarks per Inception's reporting
- OpenAI-compatible API; integrates with LiveKit, Pipecat, Vapi, and Retell
Use cases
- Enterprise AI phone and voice agents needing natural conversational latency
- Tool-calling voice assistants built on existing voice-stack frameworks (LiveKit, Pipecat, Vapi, Retell)
- Cost-sensitive, high-throughput conversational agent deployments
Pricing
$0.40 per million input tokens, $1.50 per million output tokens; 50% off introductory pricing at launch.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- Not listed
- Record updated
- 3 Oct 2026
- Record last checked
- 3 Oct 2026