Live drop

Mercury Voice

InceptionDropped Sep 29, 20265 builds5 makers2 prompts

Diffusion LLM tuned for sub-320ms first-token voice agents with tunable reasoning and tool use.

Browse in feed Submit a build

← All model drops

Live drop

Mercury Voice

InceptionDropped Sep 29, 20265 builds5 makers2 prompts

Diffusion LLM tuned for sub-320ms first-token voice agents with tunable reasoning and tool use.

Browse in feed Submit a build

About this release

What is Mercury Voice?

Mercury Voice is a diffusion-based LLM optimized for real-time voice agents. It supports low/medium/high reasoning effort, 128K context, tool calling, structured outputs, and long system prompts while achieving median time-to-first-token under 320 ms on production prompts. It is available to enterprise customers via the Inception API as an OpenAI-compatible endpoint and integrates with stacks like LiveKit, Pipecat, Vapi, and Retell.

Read the official Inception launch page

Worth opening first