Built for enterprises running customer-service voice agents that need quick first replies; general access is limited to enterprise customers.
Kenning's take
What it does
Mercury Voice is a diffusion language model from Inception tuned to run voice agents. It reasons, calls tools and follows long system prompts, and Inception reports a median of 320 milliseconds to the first answer token on real customer-service prompts. It has three reasoning settings and a 128K context, and is generally available to enterprise customers through the Inception API.
- 320 ms median time to first answer token (Inception figure)
- Low, medium and high reasoning effort
- 128K context with up to 50K output tokens
- Tool calling and long system prompts
- Tested on telecom, retail and airline agent benchmarks
Who it suits. Companies building AI phone and customer-service agents.
Key facts
- Category
- AI models
- Made by
- Inception
- Released
- 29 September 2026
- How to get it
- generally available for enterprise customers through the Inception API
- Works on
- API
- API price
- $0.40 per million input tokens, $1.50 per million output (launch offer $0.20 and $0.75)
FAQ
When was Mercury Voice released?
Inception made it generally available for enterprise customers on 29 September 2026, after a preview alongside Mercury 2.5.
How fast does Mercury Voice respond?
Inception reports a median of 320 milliseconds to the first answer token, and 750 milliseconds at the 95th percentile.
How much does Mercury Voice cost?
List price is $0.40 per million input and $1.50 per million output tokens, with 50% off at launch.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.